Hans-WilhelmEckert

Storytelling

withdata

Gaining insights, developing

strategy and taking corporate

communications toa new level

Storytelling with data

Hans-WilhelmEckert

Storytelling With

Data

Gaining Insights, Developing Strategy

AndTaking Corporate Communications

ToANew Level

ISBN 978-3-658-38554-5 ISBN 978-3-658-38555-2 (eBook)

https://doi.org/10.1007/978-3-658-38555-2

Wiesbaden GmbH, part of Springer Nature 2022

is book is a translation of the original German edition „Storytelling mit Daten“ by Eckert, Hans-

Wilhelm, published by Springer Fachmedien Wiesbaden GmbH in 2021. e translation was done with

the help of artiﬁcial intelligence (machine translation by the service DeepL.com). A subsequent human

revision was done primarily in terms of content, so that the book will read stylistically diﬀerently from a

conventional translation. Springer Nature works continuously to further the development of tools for the

production of books and on the related technologies to support the authors.

is work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether

the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of

illustrations, recitation, broadcasting, reproduction on microﬁlms or in any other physical way, and

transmission or information storage and retrieval, electronic adaptation, computer software, or by similar

or dissimilar methodology now known or hereafter developed.

e use of general descriptive names, registered names, trademarks, service marks, etc. in this publication

does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant

protective laws and regulations and therefore free for general use.

e publisher, the authors, and the editors are safe to assume that the advice and information in this book

are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or

the editors give a warranty, expressed or implied, with respect to the material contained herein or for any

errors or omissions that may have been made. e publisher remains neutral with regard to jurisdictional

claims in published maps and institutional aﬃliations.

is Springer Gabler imprint is published by the registered company Springer Fachmedien Wiesbaden

GmbH, part of Springer Nature.

e registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany

Hans-Wilhelm Eckert

Momentum Communication

Munich, Germany

Dealing with data is one of the core competencies of our digital age. And

it has long since ceased to be the domain of computer science, but needs

to build bridges to other disciplines. In his book, Hans-Wilhelm Eckert

shows the role that data plays in communication and marketing and how

it becomes an important source of storytelling.

e story is not in the data, but in our heads: this is the core message

of the book and an appeal to our own power of judgement. Data do not

have meaning in themselves, but only gain it through the context in

which we place them. In this way, the author also formulates the claim to

all those who deal with data: Don’t hide behind data, but reﬂect and

defend your point of view.

Hans-Wilhelm Eckert calls this cognitive process the oracle principle:

a 3000-year-old cultural technique that has not lost its fascination even

in the age of digitalization. On the contrary, it seems to be more eﬀective

than ever. e only diﬀerence is that today’s oracles are no longer located

in Delphi, but in Mountain View or Shenzen. And that the source of

wisdom is no longer to be found in divine inspiration, but in data.

What matters are the questions we ask of this data. is has not

changed since Delphi. Hans-Wilhelm Eckert shows this with concrete

examples from the everyday life of communication and marketing man-

agers such as the positioning of brands, the analysis of target groups and

the identiﬁcation of relevant topics. In addition to exciting insights into

Foreword

vi Foreword

narratives of artiﬁcial intelligence and the ﬁght against epidemics, the

book provides solid foundations, useful tools and examples around story-

telling with data.

And it has another good message in store: It has never been easier to

emancipate yourself from the big oracles from China and the USA. Data

is much more readily available today. Communication and marketing

managers are well advised to ﬁnd their role in the use of data in the com-

pany, to look for allies and to see the interdisciplinary interaction as an

opportunity. Digitalization gives them an important asset: access to the

customer.

With this approach, the book provides an important impetus to take

an interdisciplinary approach to the challenges of digital transformation.

is is also a concern of the Bavarian Research Institute for Digital

Transformation (bidt), which is committed to advancing digitization in

Germany for the beneﬁt of the people. is means turning digitization,

which for a long time seemed to be a purely technical matter, into a mat-

ter for everyone, because it aﬀects everyone.

Bavarian Research Institute for Digital

Transformation (BIDT), Board

of Directors ISF

Munich, Germany

November 2020

Andreas Boes

vii

e idea for this book came to me during a project report by a colleague.

At a meeting of communications managers, he told me about his goal of

predicting his clients’ issues. e company he works for covers a wide

range of consulting services. Clients expect high-quality input. It takes a

lot of time and eﬀort to create such content. But what if the elaborately

produced content doesn’t meet the client’s interest at all? Disappointed

authors, lost time and wasted money. at’s why this colleague set out to

identify the topics that mattered to his clients. More speciﬁcally, the goal

was to predict which topics would be relevant to them in the future. After

all, content production takes time. Big data and artiﬁcial intelligence

should help to support this process. Data from social media channels and

journalistic publications were used to identify relevant topic clusters and

predict their development.

is question opened the view on a wide ﬁeld of new searching move-

ments around the telling of stories. Data plays a central role in this. In

this case, data should provide information about customers’ interests and

needs. ey were not the content, but the context of the message.

But data is often also the content of messages. Here, it is a matter of

their preparation, usually in visual form – this is the classic ﬁeld of data

storytelling. In both cases, data plays an important role in gaining insights

and developing messages.

Preface

viii Preface

ey help us gain insights into people’s needs, desires and interests.

More and more even believe they can predict the future with data. at’s

a big promise. e process is similar to that of an oracle, where prophets

and priests have dedicated themselves to foreseeing the future.

A second impulse came from the side of economists: stories have an

inﬂuence on our behaviour and our economy. Robert Shiller expressed

this thesis in his book Narrative Economics. e economist was one of the

ﬁrst in his profession to analyse the economic inﬂuence of grand narra-

tives – of the gold standard, of the American dream, of the Great

Depression. I was particularly fascinated by his approach of understand-

ing the spread of stories as epidemics and applying mathematical model-

ling methods to them.

en came the real pandemic, and we all witnessed the importance

given to data in explaining the spread, and how this gave rise to a multi-

tude of readings that competed for the prerogative of explanation and

constantly evolved. Corona triggered two epidemics: the one caused by

the virus itself and the epidemic of narratives about the virus.

at was a good breeding ground for my work. A book project is

always a leap into the unknown. Of course, at the beginning, I had a

speciﬁc idea of what it should be about and why. I used this to convince

my publisher. But then new insights emerged during the research, each

time combined with the question of whether they were important enough

to change the original plan. e external developments caused by the

pandemic have once again shifted some accents. To put it brieﬂy, they

have conﬁrmed and sharpened my attitude: We are all looking at new

infections, R-values, incidences, mortality rates and many other key ﬁg-

ures. But more than the numbers, we hang on the lips of the interpreters

of these data. What’s really exciting about storytelling with data are the

prophets and priests who interpret it, and thus move others to action.

anks to digital media, we can analyse more than ever the networks that

emerge through the interaction of these stories and in which we negotiate

our view of the world.

In my writing, I have always noticed how much the historian in me

comes through. So, I have often given in to the temptation to trace the

emergence of developments such as data journalism, artiﬁcial intelligence

and data ethics, in order to understand what the protagonists’ motives

ix Preface

were and the reasons for their success. A fundamental principle of our

discipline also shapes the approach in this book: the principle of source

criticism. Data, in this sense, are sources. And with sources, a historian

always asks himself a series of critical questions, above all: why does this

source exist? What is its quality? What is its intention? is has proven to

be a very useful attitude when looking at data sources.

Of course, the book is primarily shaped by my experience in the com-

munications industry over the past 25 years: as a journalist, as a market-

ing manager, as a press oﬃcer, as an investor relations manager and now

as a communications consultant. Perhaps these diﬀerent roles already

make it clear that I don’t think too much of thinking in strict boundaries

of disciplines such as advertising, marketing and PR. e digitalization of

communication has done a lot to dissolve these boundaries and open up

new ﬁelds of activity. Much more important than the boundaries is there-

fore the question of what we can learn from the others. is could be

designers, data analysts or journalists, for example. To this end, it helps to

have an understanding of the respective roles and to focus on synergies.

With this in mind, the book is not written for any particular communi-

cations discipline, but rather to encourage us to look beyond our own

discipline and learn from others. In view of the rapid, also technologically

driven, change, I consider this indispensable.

I particularly enjoyed the excursions into other disciplines – such as

computer science, epidemiology and psychology. Of course, I am not an

expert in these subjects, and the experts from these disciplines may for-

give me for that and for my mistakes. Rather, I was also concerned here

with the question of what impulses we can take up for communication

from these subjects.

e book is aimed at practitioners from the communication disci-

plines. Due to the large number of topics addressed, it does not provide

concrete instructions for implementation, but gives impulses to deal with

interdisciplinary questions and should ideally open up new perspectives

for one’s own ﬁeld of activity.

Such a book is not possible without the experts who supported me

with their expertise and gave me valuable feedback. My thanks go to

these people who have supported me in word and deed while writing the

book. ese are in particular:

x Preface

• Katharina Brunner, data journalist

• Lutz Klaus, Marketing ROI Experts

• Martin Szugat, data thinker

A special thank you goes to my family: With my sons I was able to deepen

some technological questions. ey have also supported me in collecting,

or rather scraping, many a data series. Above all, however, I would like to

thank my wife, who accompanied me throughout the entire process of

writing this book and constantly encouraged me to sharpen the theses of

my book.

I would like to express my sincere thanks to Imke Sander. As an editor

at Springer Gabler Verlag, she supervised the editing of the work and

brought it to completion.

For reasons of better readability, I have refrained from using the lan-

guage forms masculine, feminine and diverse simultaneously in this

book. All references to persons apply equally to all genders.

I hope you ﬁnd my book stimulating and full of new insights.

Munich, Germany Hans-Wilhelm Eckert

November 2020

Graphics are an important component of storytelling with data. Digital

channels oﬀer additional features such as data download, mouseover

eﬀects, sorting options and linking. To help you take advantage of these,

I’ve put together the appropriate graphics on a website:

www.data- storyteller.de

I indicate in the captions when an interactive version of the graphic is

available.

Interactive Graphics

xii Interactive Graphics

pharmakon, Greek: poison, drug, medicine

“Sugar Man, you are the answer, that made

my questions disappear” Rodriguez, “Sugar

Man” from the album “Cold Fact”.

“Data wins arguments” Tim Campos, Facebook

CIO

xiii

1 The Pitch: What Is Storytelling with Data? 1

1.1 e Oracle Principle 1

1.2 Data Is the New Smoke 2

1.3 Prophets and Priests: Who Predicts the Future? 3

1.4 What Makes Storytelling with Data Special? 8

References 12

2 Storytelling in the Digital Age 13

2.1 From Stone Age Caves to Echo Chambers 13

2.2 Stories, Narratives and Epidemics 18

References 25

3 From the Question to the Data 27

3.1 e Brand: Why Am I Relevant? 30

3.1.1 Adidas and the Quantiﬁcation Bias 30

3.1.2 Volkswagen: Brand Communication with Big

Data 32

3.2 e Target Group: For Whom Am I Relevant? 34

3.2.1 An Online Shop Sharpens Its Customer Proﬁle 35

3.2.2 A Fashion Retailer Reinvents Itself 37

Contents

xiv Contents

3.3 e Topics: What Is Attracting? 40

3.3.1 What Advertising Does the Customer Want to

See? 40

3.3.2 Listening to What Moves the User 41

3.3.3 A Viral History of Artiﬁcial Intelligence 47

3.3.4 Context and Change of Meanings 56

3.3.5 Setting Tomorrow’s Topics Today 59

3.4 Where Silence Is Golden: Uplift Modelling 61

3.5 Towards a Data Strategy 64

References 66

4 From Data to Story 69

4.1 Data Journalism: With Wikileaks to the Breakthrough 69

4.2 Visualization: Basics, Tools and Best Practice 75

4.3 Finding the Hero in the Data 86

4.4 Saving Lives with Data: From Cholera to Corona 92

4.5 Data Storytelling Is Teamwork 98

References 102

5 Fair Play: What Counts in Data Stories 105

5.1 False Certainties and Deliberate Manipulation 106

5.2 Machine Bias 118

5.3 Ethical Issues of the Data Oracle 122

5.3.1 e Protection of Privacy 124

5.3.2 Algorithmic Accountability 127

5.3.3 Voluntary Commitments 128

5.3.4 Data Literacy 131

5.4 Less Is More: An Opportunity for Storytelling 133

References 136

The Pitch: What Is Storytelling

withData?

Abstract Man is a seeker of meaning. With his stories he organizes the

world, networks with others and constructs connections. Storytelling is

so eective because it is deeply embedded in the structures of our brain.

More and more of these stories today feed on data. Whether data is the

new oil, gold, or even plutonium: Ultimately, it is we humans who ascribe

these meanings to data. In storytelling, data has a dual role: it is the con-

tent and context of the message.

1.1 The Oracle Principle

“If you cross the Halys, you will destroy a great empire.” Such was the

saying of the oracle of Delphi to Croesus, king of Lydia. In his day it was

customary to consult an oracle before making great decisions. Previously,

Croesus himself had tested seven oracles in a kind of benchmark, and

only Pythia, the divining priestess of Delphi, had reliably provided the

correct answers. So it was natural to trust Pythia’s answer in this case as

well. Croesus then crossed the river Halys in 546B.C. and attacked the

Persian king Cyrus, who had moved ever closer to his kingdom. And in

doing so, he actually destroyed a great empire: his own.

part of Springer Nature 2022

H.-W. Eckert, Storytelling with data, https://doi.org/10.1007/978-3-658-38555-2_1

Delphi was considered the center of the world at that time. e oracle

was the most important institution of the Hellenic culture. For more

than a thousand years, the powerful sought advice from the oracle before

making important decisions. e oracle was dedicated to the god Apollo,

the god of light, purity, the arts and divination.

In direct contact with Apollo was the prophetess of the oracle, called

Pythia. Only the priests had contact with her. Pythia sat in the holy of

holies of the temple on a tripod above an opening in the oor, through

which she could enter into contact with Apollo according to the view of

that time. e sentences uttered by the prophetess Pythia were recorded,

processed, interpreted and then given to the questioner as a prophecy

(about the oracle of Delphi see Maaß 1996).

e oracle was known to provide answers that had multiple readings.

e intention of the questioner was directed into the future. e answers,

in turn, were often only revealed in retrospect, and until then they often

misled the questioner. ey sent the hero on a journey in the course of

which he had to win battles and at the end of which, in the best case, he

recognized himself. is was not always pleasant. Neither with Croesus

nor with Pyrrhus or Oedipus. For that, the heroes’ journey provided the

dramas that make up Greek mythology. And not only them. e inven-

tion and telling of stories is a basic principle of our human existence. And

the heroes’ journey became the much-cited formula of storytelling.

e oracle principle forms the blueprint of this book: the art is not

only to ask the right questions to an appropriate addressee, but rather to

then interpret the answers received correctly. e span between the for-

mulation of the question, i.e. the inner departure of a hero, and the

insight, i.e. the end, is the drama of the journey that is at the core of what

we now call storytelling.

1.2 Data Is theNew Smoke

What has not already been attributed to data? “Data is the new oil” was

the headline of the Economist in 2017, referring to the new driving force

of the fourth industrial revolution (Economist 2017). “Data is the new

plutonium,” said Jim Balsillie, founder of Blackberry, referring to the

H.-W. Eckert

toxic content of the material from which all modern dystopias are fed

(Balsillie 2018). “Data is the new gold,” says a study by Accenture, which

believes that as democratization continues, data will be the central driver

of growth for every company (Accenture 2020).

e thesis of this book is: data is the new smoke. e source of wisdom

today is no longer Apollo, but it is data. And at the heart of the cult of

data today are no longer prophets and priests, but people who know how

to distill information from data and those who use it to make predictions

and assert their claim to interpretation.

Data is the smoke that pours from the crevices of our oracles. Everything

is contained in this smoke. Whether it is fuel, poison or gold, whether it

harms or helps us, sobers or intoxicates us, that is up to us. Only the

Pythia and the priests make predictions from it that give direction to our

actions. ey decide signal or noise. ey weave the stories from it, they

create what we call meaning from the data. And we all, as oracle receivers,

have the freedom to draw our own conclusions from these oracle sayings

and base our actions on them. Even though today we are surrendering

more and more of this interpretive power to so-called intelligent machines,

it is still us who have decided to do so. ese machines have as much

power over us as we are willing to give them. It is in our hands to do so.

In that sense, this book is also an appeal to our own power of judgment.

1.3 Prophets andPriests: Who Predicts

theFuture?

I worked in the nancial industry for many years, including several years

in investment banking. During this time I was able to observe the system

of companies, stock exchanges, banks and investors from the inside. A lot

of money was made in the capital market for a long time and some still

make it in these times. is led to the development of a division of labor

around the question of how money will be worth more in the future,

consisting of issuers, brokers, asset managers, analysts, private and insti-

tutional investors and the media, who fuel the system with their stories.

1 The Pitch: What Is Storytelling with Data?

For the capital market, the data oracle is constitutive: CFOs, together

with their investor relations departments, develop the story of their com-

pany in order to achieve the best possible price for their company shares

on the markets. To interpret these stories, there are analysts on the sellers’

and buyers’ sides who formulate their own readings of these stories, at the

end of which the oracle says: “Target price 435 euros”. e negotiation of

these stories takes place at the stock exchanges (and of course at many

other trading places), where we can follow the outcome of the negotia-

tion every second.

e big data reservoirs here are Bloomberg and

Reuters, which pay dearly for access to the most precious data and there-

fore secure an information advantage for those who can aord it.

Similar structures can also be found in other industries. Anyone look-

ing for current technology trends will not be able to avoid the analyst

rm Gartner and its hype cycle (Fig. 1.1). e hype cycle depicts the

phases of public attention through which a new technology passes: from

Here I am only talking about the visible area. e ash boys of high-frequency trading are another

issue that falls more into the realm of piracy.

Fig. 1.1 The Gartner Hype Cycle: every year the market researcher updates its

cycle of current technology trends. (Source: Gartner)

H.-W. Eckert

its technology trigger, through the peak of inated expectations, to the

trough of disillusionment and nally to the plateau of productivity. e

image of the hype cycle was coined by Gartner consultant Jackie Fenn

and still forms the company’s brand essence today. Every year, new assess-

ments of current technologies are eagerly awaited: Where are new hypes

emerging, which technology is on its way to the trough of disillusion-

ment, and where are real productivity gains? A marketing machine driven

by annual updates feeds o the basic narrative embodied by the curved

curve of the hype cycle. It’s hard to build a business model on a story bet-

ter than that.

What Bloomberg and Reuters are to the capital markets, market

research institutes such as Nielsen, Ipsos, IQVIA and Kantar are to the

economy. ey deliver relevant studies for almost every question. In the

eld of communications, the major market media studies should be men-

tioned here, which conduct representative research into media use and

consumer behavior on the basis of large case numbers. Market researchers

such as the Sinus Institute access this data material directly and oer

individual counts and analyses. Evaluations are possible, for example,

according to various markets such as nutrition, nance or travel, and

oer the possibility of evaluating them according to socio-demographic

characteristics and also according to the so-called Sinus Milieus.

e days of these oracles and many other business models based on

this pattern are numbered. e sources of our narratives are and will

remain data. But the way data is collected and distributed is about to

undergo a massive change. e business model of analyst rms and mar-

ket researchers is based on the fact that they have exclusive access to data.

ey have a pool of market data, company data, supplemented by pri-

mary and secondary surveys. e collection, processing and preparation

of the data is costly, requires in-depth methodological knowledge and for

many years secured the companies an interpretative edge on which their

business models are based.

But the business model of these data giants is dissolving. e quality

and quantity of the available data are completely dierent today. We have

long since tapped into data sources that let us make decisions faster and

nd new insights. e internet, social media and the cloud make data

available in real time. Company gures, for example, can be

1 The Pitch: What Is Storytelling with Data?

automatically extracted from press releases, investor presentations and

annual reports at the time of publication. Today, hardly any humans are

needed for this. A well-programmed algorithm works faster, more pre-

cisely and around the clock.

Even more signicant than the Internet itself is a piece of hardware:

the rst iPhone. In 2007, Apple launched the mother of all smartphones.

At the time, perhaps not even Steve Jobs had any idea how much it would

change our entire lives. ings we previously owned materially disap-

peared into it: rst our address books, then our camera, music collection,

wallet. Not only did they disappear into it, but they miraculously com-

bined to create new things and generated new data and metadata, data

for organizing and classifying data. WhatsApp doesn’t know what we say,

but it knows when, where and with whom we speak. e photos on our

smartphone store time and place, which then reappear in our Time and

Travel line as memories, as well as on Facebook, Instagram and Co. where

we posted them. ese and many other useful or even just nice features

make the smartphone so powerful that we have adjusted our behavior to

it and are busily generating even more data and metadata.

Social media and the movement data from our mobile devices are con-

stantly providing new amounts of data. Taught by us humans– for exam-

ple when “liking” a post or commenting on a picture– machines are now

increasingly capable of understanding what we say.

In addition to the consumer market, industrial applications are ood-

ing much more data onto the market or into the cloud: sensors in cars,

robots and entire industrial plants generate many times more data and

feed cloud systems in which it is provided and processed. Estimates from

the industry analyst IDC from 2019 assume an almost unimaginable

amount of over 40 billion networked devices in 2025 (IDC 2019).

If data is the source of our divination, then these devices provide the

raw material that feeds our data oracles, their prophets and priests. ey

drive all modern business models, they create new oligopolies and make

Facebook, Google, Amazon and Co. so valuable. What this means for the

oracle market is that a new Delphi League owns and uses the knowledge

of humanity: ese are the GAFAs (Google, Amazon, Facebook and

Apple) in the Western world and the BATXs (Baidu, Alibaba, Tencent

and Xiaumi) in the Eastern world.

H.-W. Eckert

But data has long since ceased to be generated solely in the system of

the major platforms. erefore, companies are well advised to build up

their own data sovereignty in order to develop a model and control com-

munication activities from it. Market researcher Forrester predicts that

brands that own their data will outperform those that merely obtain their

data from third parties. To achieve this, it is important to make oneself

independent of the large data networks and to feed knowledge about

customers from one’s own sources. Access to data has become increas-

ingly easy, allowing any company to build its own oracle. Many compa-

nies have taken this step and created technological structures and built

departments dedicated to developing questions, collecting, processing

and interpreting data. In this way, they gain their own access to a valuable

raw material and secure an edge over the competition.

Business

Intelligence is the name of the discipline that deals with all forms of data

analysis. And where it is already more about predicting the future, it is

also referred to as business analytics.

However, data in itself is not relevant. ey only become valuable

when we give them meaning. Today, we often do this in the belief that

data is the source of objective truth. Wherever data are the sources of

wisdom, there needs to be a Pythia who opens up access to these sources

and a priest who interprets the words of the Pythia. In today’s corporate

world, multiple functions vie for this interpretive authority.

Communications and marketing managers are not always ahead of the

game. But their access to the customer gives them an advantage that they

should use protably. e better companies understand customer behav-

ior and draw the right conclusions from it, the more loyal customers will

be to the company in the long run. erefore, they should have good

access to the oracle.

For example, brick-and-mortar retail has recognized how important a direct line to the customer

is for business success. e Otto Group also makes its concentrated knowledge of 25 million CRM

data in 150 segments available to third parties, see Zimmer (2019).

A good overview of these and related terms such as predictive analytics and predictive mainte-

nance is provided by Mauerer (2017).

1 The Pitch: What Is Storytelling with Data?

1.4 What Makes Storytelling

withData Special?

Storytelling is so eective because it is deeply embedded in the structures

of our brain. It is based on our tendency to make decisions emotionally

in order to justify them rationally afterwards. e psychologist Daniel

Kahneman described the phenomenon in his book “inking, Fast and

Slow” with many examples (Kahneman 2011). Storytelling approaches

use this way of working of our brain and rely on the emotional appeal of

the counterpart. ey are based on the pattern of the hero’s journey, the

drama and its resolution. is describes very well the formal side of the

phenomenon. However, this approach focuses only on a few formats,

namely reportage, commercials and other subjective stories and accounts

in which such personalization can take place. But what about analyses,

studies, commentaries news, infographics or posts? ey too take up

themes and explicitly or implicitly refer to stories and their heroes.

It therefore needs a broader view of the storytelling environment. e

focus of storytelling lies on the second component of the word and

explores the question of how we tell stories. Much has been written since

Aristotle about the construction of stories and their structure, i.e. the

formal side of the subject. I go into this in more detail in Sect. 1.2. e

what, the content and its context, its development, transformation and

interaction with other stories, is rarely addressed in storytelling.

e term “narrative”, borrowed from English, is now also used in

German in a theory-heavy or at least meaning-heavy way, for example

when it comes to the big topics such as nationalism, globalisation, the

welfare state or growth. A buzzword, perhaps, but one with the claim to

dock with postmodern narrative theory. More on this in Sect. 1.2.

For the purposes of our question and the rest of this book, I take nar-

rative to mean the context of the story being told. is context is essential

for our understanding of the world and the story we have just told. For

man is a seeker of meaning. With his stories he organizes the world, net-

works with others and constructs contexts. rough stories he communi-

cates knowledge and forms networks. ese stories are interwoven,

through them we interact. People with similar backgrounds, attitudes,

H.-W. Eckert

and desires create the primary resonant spaces of these stories. Depending

on the mix, the stories are constantly changing and can go viral far beyond

these resonant spaces. With such a network concept, the relevance and

spread of stories can be mapped and interactions can be tracked. Via digi-

tal channels, their infection paths become traceable.

So when we speak of stories here, we mean the specically told story.

Narratives,

on the other hand, are the “larger” stories that create mean-

ing and order and are agreed upon and shared by social groups. ey

provide the context of the specic story.

Based on this consideration of narratives (the macro-level with the

context of the story told) and storytelling (the micro-level with the for-

mal structure of the story told), the vertical dimension of the playing eld

emerges. On the left is the context level, on the right the actual narra-

tive level.

e second, horizontal dimension is provided by the data themselves:

e way it is prepared diers in whether it has an explanatory or explor-

atory function. If the data is prepared in such a way that patterns become

apparent, but no solution is presented yet, the task of exploring and

drawing conclusions lies with the viewer. Dashboards are built this way.

ey provide patterns from which experts can derive explanations. Visual

presentations of complex processes, the visualization of user ows or the

color clustering of medical images are also based on these methods.

Exploratory processing is the preliminary stage for the explanatory pro-

cessing of data. In this book, I will use the examples of cholera and covid

to show how exploratory and narrative processing dier (see Sect. 1.4).

When data provides explanations, it has a narrative structure. is is

the real playing eld of data storytelling– outlined in orange here.

• Process of cognition: Explaining and exploring dier in who does the

work– sender or receiver. In exploring, the receiver does the work of

e term “narratives” may have come into German usage through Lyotard from the French “recit”

via English, cf. Heine (2016): “e Oxford English Dictionary explicitly names Lyotard as the

originator of the latest English meaning of narrative, dening it: ‘a narrative or account used to

explain or justify a society or historical period’. One is therefore probably not wrong in assuming

that the conjuncture of the noun narrative in German goes back to the inuence of English.”

1 The Pitch: What Is Storytelling with Data?

interpreting and making sense, in explaining, the sender provides

the insight.

• Narrative level: On which level does the narrative take place? Does it

provide context or is it itself the story? In terms of data: Does the data

provide insights into which stories are relevant or is it itself the subject?

• Data provides the narrative: by segmenting customers, identifying

needs and themes, developing brand stories. is forms the strategic

basis for all stories that are told. Data provides the context for drafting

and spreading stories. It provides us with deep insights into the world

of our fellow human beings, their needs, expectations and desires.

• Data delivers the story: Data is the content of the story. is is the

domain of data journalism and all journalistically data-inspired for-

mats such as content marketing and social media campaigns, but also

analyst reports. Visualizations are the guiding discipline of these stories.

• Data provides the insight: Humans are masters of pattern recogni-

tion. Visualizations help to recognize and analyze correlations. Data

provides the patterns, the human or/and an AI interprets them and

derives an insight. is can form the basis for a narrative and a story.

• Data is the issue: recognizing anomalies, monitoring systems: visually

prepared data enables the control of processes, workows and states.

Dashboards, i.e. overviews of data categories considered relevant, are

structured according to this principle.

But as soon as data provide explanations, they have a narrative structure.

is is the real playing eld of data storytelling (Fig.1.2). With the pro-

cessing, analysis, interpretation and embedding in the context of existing

narratives, we humans are able to tell relevant, activating stories from

them and to develop contexts of meaning.

e journey in this book begins with a look at the literature on story-

telling and narratives. e focus here is on the question of what digitali-

zation has actually changed and what remains as communication patterns

(Chap. 2).

e topic of Chap. 3 “From the Question to the Data” uses examples

to show the contribution that data makes to the development of narra-

tives: Since we are talking about corporate communication, the journey

starts with the brand. e brand embodies the personality of a company

H.-W. Eckert

Fig. 1.2 Data storytelling– the playing ﬁeld. (Source: Own illustration)

or product and ideally answers the “why?”. As there is no brand without

a market, in the next step we look at how customers can be classied

(“For whom?”) and with which topics they can be addressed (“How?”).

Once the context is claried, it is a matter of developing the specic

stories. is is the focus of Chap. 4 “From Data to Story”. What can we

learn from data journalism, how can stories be visualized and what are

the roles in the oracle team?

Every game needs rules. Chap. 5 “Fair Play: What Counts in Data

Stories” deals with the ethical framework and the sovereign handling of

data: Around false certainties, deliberate manipulations, and what data

literacy means for storytelling.

e more the raw material of our forecasts consists of data, the more

important it is that we not only ask the right questions, but that we are

also aware of what assumptions are contained in our question. Otherwise,

in the frenzy of oversupply, we fall into the illusion that the answer is

already in the data. e Croesus case shows how prone to error we

humans are when it comes to interpreting information. And this doesn’t

just aect those who base their actions on the oracle. Even the oracle itself

can err if it is not aware of the nature of its sources.

1 The Pitch: What Is Storytelling with Data?

References

Accenture (2020) e human impact of data literacy. Accenture. https://www.

accenture.com/_acnmedia/PDF- 115/Accenture- Human- Impact- Data-

Literacy- Latest.pdf. Accessed 11 Juni 2020

Balsillie J (2018) Data is not the new oil– it’s the new plutonium. Financial

Post, 18. Mai. https://business.nancialpost.com/technology/jim- balsillie-

data- is- not- the- new- oil- its- the- new- plutonium. Accessed 22 Jan. 2020

Economist (2017) e world’s most valuable resource is no longer oil, but data

(06.05.17). https://www.economist.com/leaders/2017/05/06/the- worlds-

most- valuable- resource- is- no- longer- oil- but- data. Accessed 26 Nov. 2019

Heine M (2016) Hinz und Kunz schwafeln heutzutage vom „Narrativ“. Die

Welt, 13. November. https://www.welt.de/debatte/kommentare/arti-

cle159450529/Hinz- und- Kunz- schwafeln- heutzutage- vom- Narrativ.html.

Accessed 4 Juli 2020

IDC (2019) e Growth in Connected IoT Devices Is Expected to Generate

79.4ZB of Data in 2025, in IDC 18.06.2019. https://www.idc.com/getdoc.

jsp?containerId=prUS45213219. Accessed 2. Mai 2020

Kahneman D (2011) inking, fast and slow. Penguin, London

Maaß M (1996) Delphi– Orakel am Nabel der Welt. orbecke, Sigmaringen

Mauerer J (2017) Was ist was bei predictive analytics? Computerwoche, 11.

Dezember. https://www.computerwoche.de/a/was- ist- was- bei- predictive-

analytics,3098583,5. Accessed 28 Mai 2020

Zimmer F (2019) Daten, zu mir!– Wie Marken sich datentechnisch emanzip-

ieren Dmexco, 8. November. https://dmexco.com/de/stories/daten- zu- mir-

wie- marken- sich- datentechnisch- emanzipieren/. Accessed 25 Jan. 2020

H.-W. Eckert

Storytelling intheDigital Age

Abstract Stories shape our view of the world. Innovations such as print-

ing, newspapers, radio, television and now digital channels have made it

possible to spread stories and connect humanity on a global scale.

Increasingly today, we are putting control over the selection and spread-

ing stories into the hands of machines whose actions are fed by data. We

are also unlocking new content through data: rough it, we are able to

venture into regions where us humans lack an awareness. We recognize

connections and patterns that we would not have seen without this data.

By evaluating and contextualizing this data through language, we expand

our world.

2.1 From Stone Age Caves toEcho Chambers

Somewhat human, somewhat animal: Six creatures with ropes and spears

are chasing a huge bovine. With their dark red pigments on the bare wall,

the paintings look a bit like grati. ey are the oldest hunting scenes in

the world in the cave Leang Bulu Sipong 4 in the southwest of the

Indonesian island of Sulawesi. ey tell an ancient story of hunting and

part of Springer Nature 2022

H.-W. Eckert, Storytelling with data, https://doi.org/10.1007/978-3-658-38555-2_2

killing. Comparable depictions were also created somewhat later in

Europe in the cave Cueva de El Castillo near Puente Viesgo in

Cantabria, Spain.

What makes these paintings so signicant is the artists’ ability to create

things that do not exist (Callaway 2019). Whoever immortalized himself

or herself there: He or she left behind one of the rst records of human

creativity. e paintings in the caves are the rst known sign that our

ancestors began to develop ideas and communicate about them. Even if

today we do not know what our ancestors told each other about it.

Researchers assume that language developed at the same time as painting

(Kuke 2012).

Storytelling became a need like only food, sleep, sex, or companion-

ship. “Storytelling, however, was not a pastime, it was training. It allowed

people to theorize about what the other might be up to, who was good

friends with whom, or who might just be putting on an act.” (Siefer 2015).

Storytelling is as old as humanity. In a community it was conducive to

be cooperative and helpful, to be able to read emotions and react to them.

is helped people to stand their ground and stay on top of things. e

means to do this was through stories. When our ancestors began to talk

around the campre about things that existed only in their imaginations,

they began to network and organize.

Storytelling has been deeply inscribed in the structures of our brain in

the course of evolution. Neuroscientists have studied how stories work in

our brains and have come to impressive conclusions. Uri Hasson vividly

explained the eects of stories on our brain in a TED Talk (Hasson 2016).

• Neurophysiologically, stories are “cinema in the head”: they activate

dierent areas of our brain. In addition to the language centre, which

is located in Broca’s and Wernicke’s areas in the cerebral cortex, areas

in the insular cortex (cortex insula) are also activated, an area that is

responsible for feelings of empathy, pain and pleasure.

• Stories stimulate our brain to release hormones that are important for

feeling stress or empathy, for example.

• With storytelling, stories synchronize the brains of the teller and the

listener. is is how we appropriate stories and take them for our own

experiences.

H.-W. Eckert

• Because of the powerful eects on our brains, stories can be remem-

bered up to 22 times better than pure facts, says Jerome Bruner,

Harvard professor and co-founder of cognitive psychology in his work

“Actual Minds, Possible Worlds” (Bruner 1986).

is is how patterns developed around the campres of our ancestors and

shaped our narrative structures– from the rst cave pictures to posts on

Instagram. Images were likely the foundation of oral narratives.

Subsequently, it was always new media that people used to extend the

reach of their stories. After the introduction of writing, it was the inven-

tion of the printing press that made the book the rst mass medium. In

his country, Martin Luther helped the technology to achieve a break-

through by translating the New Testament into German in 1521, giving

a large audience immediate access to the Scriptures.

On the basis of language, all the great narratives that dene our lives

have emerged: Religions, philosophies, ideologies and political theories.

Even in the so-called exact sciences, the social consensus of a group of

scientists determines which view prevails. In his 1968 analysis of the

structure of scientic revolutions, omas Kuhn showed how in science

the wrong is not replaced by the right, but only one agreement is replaced

by another, and coined the term paradigm shift for this.

After all, all the institutions that shape our world have also developed

from these narratives. States, churches, and schools, for example, have

emerged from stories that people share, negotiate, agree upon, and on the

basis of which they eventually build the organizational structures. In this

way, we have created our powerful systems that dene our world. Whether

it is science, religion, politics or economics, in the end it is all about

asserting the interpretive sovereignty of social groups and thus exercising

power. And where power is concerned, manipulation is not far away.

Media play a central role in this. Innovations such as printing, newspa-

pers, radio, television and now digital channels have contributed deci-

sively to the emergence of systems of order and the networking of

humanity. e more powerful a medium became, the stronger the pos-

sibility of manipulation.

One certainly signicant dierence from analogue times is that today

we have put the control for selecting and distributing stories into the

2 Storytelling in the Digital Age

Fig. 2.1 The term ﬁlter bubble was popularized by internet activist Eli Pariser

(2011) and surpassed the term echo chamber in late 2015. Interactive graphic at

www.data- storyteller.de. (Source: Google Ngram Corpus German 2019)

hands of machines. Algorithms decide what we see, bots spread stories. A

shadow industry oods the world with fake news

and so-called deep-

fakes make it harder to identify fakes today. But most importantly, much

of the channel is in the hands of a very small number of companies. is

is worrying.

Regarding the growing power of Google and Facebook, the Internet

activist Eli Pariser (2011) expressed his criticism with the image of the

lter bubble (Fig. 2.1). e large Internet corporations would exploit

knowledge about their users to bring together those who have similar

interests. A lter eect controlled by algorithms leads to my Facebook

timeline, for example, only showing me posts that conrm my own opin-

ion. In this way, Pariser’s allegation goes, the algorithms of internet

See, for example, the great report by Samanth Subramanian on the city of Veles in Macedonia,

which has become the epicenter of the fake news industry in the Trump election campaign. On the

Russian disinformation industry, see the website of the EU East StratCom Task Force with a weekly

newsletter on the latest fake news at https://euvsdisinfo.eu/

H.-W. Eckert

corporations further reinforce opinions in so-called lter bubbles and

thus further fuel extreme positions.

In the discussion triggered by it, his

thesis was in particular attacked on two points: On the one hand, manip-

ulation was not as pronounced as Pariser implied; on the other hand,

given the ood of information, a ltering function of all media was

unavoidable.

An internal Facebook investigation, recently reported by

the Wall Street Journal, proves that Facebook was well aware of how

much its algorithms contributed to the polarization of its users. e nd-

ings were inconsequential, however, and the study ended up in the com-

pany’s archives (Horvitz and Seetharaman 2020).

Furthermore, the accusation remains that the user himself has no

inuence on what is fed to him by the algorithm and has no insight into

how this decision was made.

e related theory of echo chambers is also based on the assumption

that interaction with like-minded people in social networks leads to a

fragmentation and narrowing of the world view. However, it is broader in

scope and does not focus solely on the inuence of algorithms. It is based

on a bias we call conrmation bias. In psychology, this describes the ten-

dency to select and interpret information in such a way that it corre-

sponds to one’s own wishes.

e conrmation error, by the way, also sealed the fate of Croesus

mentioned at the beginning. e message of the oracle was: “You will

destroy a great empire.” In his wishful thinking Croesus perceived only

one possibility of interpretation, namely, that it must be the empire of his

adversary. at it might be his own empire he did not consider. Yet the

oracle itself warned of the eects of his sayings. At the entrance of the

temple there is said to have been an inscription, “Know thyself.” Perhaps

Croesus should have thought about this inscription before he crossed

the Halys.

e multiplicity of digital channels allows us to avoid self-awareness

and surround ourselves with the people (or machines, it is not always

Here I simply assume that Facebook is also a medium, even if Marc Zuckerberg stubbornly denies

this. On the lter bubble, see Pariser (2011).

See Seemann (2018) for the current debate.

e theory for this comes from Peter Wason and Philip Johnson-Laird (1968).

2 Storytelling in the Digital Age

certain) who share our opinions. In this sense, the Internet has only mul-

tiplied the caves of our ancestors. e basic pattern of communities form-

ing by inventing and spreading stories hasn’t changed for 40,000years.

It’s just that today we have the freedom to change caves if we don’t like

the narrative that prevails there.

2.2 Stories, Narratives andEpidemics

e literature around storytelling focuses on a recurring pattern: the

hero’s journey. It is explored in countless facets.

In essence, they can be reduced to a few essential elements:

e Hero: A single character is the protagonist of the story. is personal-

ization provides the reader with the opportunity to identify with him.

From there, the connection to a larger narrative develops. e journey of

the hero, who leaves his familiar surroundings and sets out into a new,

unknown world. Here, various stations await him: adventures, challenges,

and tests of endurance that he grows from– physically or even as a person-

ality. Until he nally has to pass the decisive test. Once he has successfully

completed this, the reward awaits him. From there he returns home as a

new person, is celebrated and reports on the change (Campbell 2008,

pp.151–206).

e essence of any good story is based on such pattern. But a hero rarely

comes alone. It takes personnel to animate a story. is is where the

antagonist comes in: the anti-hero, the villain, or however the role is

eshed out. e stronger the antagonists, the greater the fall, the more

drama the story oers. e Beauty and the Beast, Dr. Jekyll and Mr.

Hyde, Faust and Mephisto. All the other variations can be derived from

this basic constellation: Allies and traitors, converts and renegades. e

most important ingredient of a good story is the resolution of a conict.

is has been described extensively in the literature. Anyone who

wants to sharpen their pen as a screenwriter, for example, will not be able

to avoid the standard works by Joseph Campbell and Robert McKee.

Campbell spans millennia of cultures and discovers recurring narrative

H.-W. Eckert

patterns (what he calls monomyths) that can be found in Greek mythol-

ogy, Siegmund Freud’s interpretation of dreams, and Carl Gustav Jung’s

teaching of archetypes, among others. George Lucas was inspired by

Campbell’s work to develop his Star Wars saga. Robert McKee also

focuses on the big themes and introduces the art of screenwriting for the

Hollywood stu. For a heroic epic, such an approach certainly works

well. To captivate my audience for 21h and 48min (that’s how long all

Star Wars episodes combined last), I can explore many facets of my hero’s

struggle against evil. is is rarely the case in corporate communications,

where the issues are usually not big enough to dock onto the basic ques-

tions of life. As readable and inspiring as these works are: they are of

limited use as guides to our concerns. Other works are more helpful for

communicators: “Storytelling” by Petra Sammer and “ink Content”

by Miriam Löer provide very good basics on storytelling for companies

(Sammer 2017; Löer 2014).

at a picture is worth a thousand words is a common stereotype. Our

eye is a master of pattern recognition. is already helped our ancestors,

for example when they distinguished edible from poisonous mushrooms

or discovered the outline of a sabre-toothed tiger in the savannah in time.

Our optic nerve forms the shortest and fastest path of all sensory organs

to the brain. at’s why visualizations are so eective. And so it is not

surprising that a whole branch of storytelling deals with it. For example,

the book “Visual Storytelling” by Pia Kleine Wieskamp, of which, how-

ever, only a small part of the book also deals with infographics, i.e. the

data-based part. Nancy Duarte uses data storytelling in the title of her

new book. Unfortunately, the result is only a guide on how to create

trenchant presentations for the busy CEO (Kleine Wieskamp 2019;

Duarte 2019). A systematic treatment of storytelling with data is pro-

vided by Brent Dykes in his 2020 work Eective Data Storytelling. He

precisely analyzes the specics of data stories and makes clear in which

form the narrative structures of a hero’s journey can be implemented here

as well. He sensitizes to perceptual distortions and shows which visualiza-

tions are suitable for data stories. It is the most profound work on the

subject, which is why we will encounter it again and again in this book.

At this point, it is also worth looking at developments in data journalism,

which has taken o with the processing of big data from WikiLeaks and

2 Storytelling in the Digital Age

has not only set standards for the analysis and processing of big data, but

can also serve as a template for collaboration between interdisciplinary

teams. More on this in Chap. 3.

Humans are pattern seekers, says Daniel Kahneman in his book quoted

earlier. We believe in a coherent world in which all things make sense and

events connect causally. People recognize patterns even in things that are

statistically random (Kahneman 2011, pp. 114–118). Storytelling

describes the formal side of this pattern recognition. By inventing and

telling stories, we are able to structure and interpret our world and neu-

rally couple ourselves with other people. In a sense, stories are the data

packets in which the myriad of impressions pelting us are ltered, com-

pressed, and connected to meaning. anks to powerful media, we can

use these stories to synchronize very large numbers of people in their

actions over very large distances. According to this understanding, story-

telling is nothing more than data processing for people.

But the approach still falls short for the following consideration. It

needs a look at the narratives, the content and context of the many sto-

ries. For only those who have sovereignty over the interpretation of the

stories also have power over the people.

e oracle provides the paradigm here. Today there is no longer a clear

centre like the one at Delphi, which provides us with a binding interpreta-

tion and an organising principle for all stories. e world appears frag-

mented. In our post-industrial society, the certainties of modernity have

also been destroyed. e idea of the Enlightenment, with its goal of the

self-liberation of the individual, has proven to be a mistake. is is how

Jean-Francois Lyotard, a French philosopher and literary theorist, put it in

his 1979 work La Condition Postmoderne. e narratives of modernity,

he argues, began with the claim to possess a scientic legitimacy. Lyotard

declares this narrative to have failed. Our postmodern world is character-

ized by the coexistence of dierent narratives, which Lyotard calls dis-

courses. ese discourses follow their own rules as isolated language games

between which there is no understanding and no interaction (Lyotard 1979).

Still believing in the triumph of liberalism as the dominant interpreta-

tion of the world, the American political scientist Francis Fukuyama

argued in his 1992 book e End of History that the world had a prede-

termined path and that economic and political liberalism had won the

H.-W. Eckert

day. With the fall of the Berlin Wall and the collapse of the Soviet Union,

the last grand narrative competing with the Western world– Marxism–

had eectively failed. Western democracy and liberalism had prevailed

and thus achieved the goal of history (Fukuyama 1992). is reading

proved to be a fatal error with the 2001 attacks on the World Trade

Center and the Pentagon.

e American military already saw things dierently in the 1990s. e

acronym VUCA was used to describe a world that had lost its old certain-

ties and was characterised by volatility, uncertainty, complexity and

ambiguity. e concept was developed at the United States Army War

College (USAWC) and initially served to describe the multilateral world

after the end of the Cold War and to develop strategies for asymmetric

warfare. Today, the term plays an important role in the teachings of stra-

tegic leadership and organizational theory, particularly in the context of

digitalization and agile leadership (for its origins, see U.S.Army Heritage

and Education Centre 2019). Most importantly, the concept is another

expression of a fragmented mishmash of narratives in which many oracles

compete for interpretive authority.

Stories do not even need to be true in order to have an eect. What is

decisive is that a community shares and spreads these stories (and thus

also conceals other stories). It is by no means new that this involves delib-

erate falsication in order to put things in the right light. In the face of

some excited fake news, it helps to remember that stories have always

been invented to achieve goals. Fakes were also deliberately commis-

sioned in the past, for example to prove ownership claims or to legitimise

ruling houses. e Privilegium Maius of 1358/1359, for example, is one

of the most skilful document forgeries of the Middle Ages. With its help,

the Habsburgs legitimized the state right to dominate the Austrian lands.

ese forgeries and manipulations have left their mark on human history.

Historian Eric Hobsbawn, in an edited volume, has called these narra-

tives the “Invention of Tradition,” noting that many so-called traditions

that we think have endured for a long time have been recently invented

(Hobsbawn and Ranger 1983). And another historian, Yuval Noah

Harari, refers to man as a post-factual being who has only been driven,

forming his networks, and gaining power over the world by believing in

ctions (Harari 2018, pp.231–244).

2 Storytelling in the Digital Age

ere are no longer any binding narratives in our society. Our knowl-

edge consists of a multiplicity of narratives that coexist. To this extent,

postmodern Lyotard’s ideas seem to coincide with the masterminds of the

American military academy and the media theorists with their echo

chambers. But rather than persisting in isolated resonant spaces, some

scholars view narratives as large, interactive systems. Stories would inter-

act with each other as if on a large playing eld– reinforcing each other,

ghting each other, and constantly modifying each other. at’s the

model economist Robert Shiller develops in his book Narrative

Economics. He shows that these stories actually also have real inuence

on our actions and thus on economic development – an observation

neglected by economists for far too long. Shiller demonstrates the eect

of various popular narratives on economic development in the US.

Shiller has described their enforcement in terms of epidemics: e tri-

umphant march of a story behaves much like the course of an Ebola

infection (e book is written before Corona). e infection curve ini-

tially rises as more people get infected than recover or die. is process

reverses once the epidemic has passed its peak, i.e. the number of new

infections decreases relative to those who die or recover. Overall, the

hump-shaped course typical of epidemics is thus formed. According to

Shiller, the same applies to the rate of infection of narratives. Here, simi-

lar to viruses, there are narratives that are more infectious than others.

“Similarly, with narrative epidemics there may be two dierent narra-

tives, one with some minor story details that make it more contagious

than the other. e minor story details make the rst narrative, and not

the second, into an epidemic” (Shiller 2019, pp.18–21, quoting p.21).

Shiller examines the impact of these stories using common economic

narratives– such as the gold standard, the Great Depression and Bitcoin.

Its truthfulness plays no role in the enforcement of the story:

“Ultimately, a story’s contagion rate is unaected by its underlying truth.

A contagious story is one that quickly grabs the attention of and makes

an impression on another person, whether that story is true or not”

(Shiller 2019, p.96). Stories unfold their impact when they appear coher-

ent and help to create meaning by linking up with other stories. In many

cases, stories are in circulation long before they become infectious. is is

well illustrated by the American Dream narrative, for example. First

H.-W. Eckert

circulated in James Truslow Adams’ 1931 bestseller e Epic of America,

Shiller sees the narrative as an example of a slow epidemic that is still

relevant and growing today. From Adams’s perspective, the American

Dream was the dream of a social order in which everyone would nd a

place according to his or her ability that would oer recognition and

livelihood, regardless of birth or social position. Adams thus refers to the

American Declaration of Independence and the principle of equality for-

mulated there.

But for the narrative of the American Dream to go viral, it needed

more triggers. One of them was Martin Luther King’s speech in 1963.

Here we see the main three building blocks that helped to signicantly

increase the infection rate:

• a famous protagonist: in this case in the person of the civil rights activ-

ist Martin Luther King

• a large audience: his speech “I have a Dream” on 28 August 1963in

front of the Lincoln Memorial in Washington

• a massive conict: racial discrimination and the demand for equality

of blacks and whites

It is only in this constellation that the story goes viral and provides the

impetus for the further infection cycle of the American Dream narrative,

which continues into the present. us, this narrative is also a very good

example of how long infection cycles can last (Shiller 2019, pp.151–154).

Taking this approach further, one of the most important principles of

storytelling is to form topics and messages in such a way that they can

dock onto already familiar themes and messages. Ideas, topics and atti-

tudes shared by a group of people promote social cohesion. us, narra-

tives form something like the social glue of a community.

In times of the Corona pandemic, however, some communication

managers will ask themselves whether it is still an appropriate metaphor

to infect people with stories. e asymmetrical communication model

with the clear distribution of roles between sender (communication

driver) and receiver ((target) customer) has become obsolete at the latest

with the triumph of social media platforms. It has long been accepted in

communication science that the roles of the receiver of content have

2 Storytelling in the Digital Age

shifted to the creators, distributors and commentators of content.

rough digital channels, we now also have the data to trace the emer-

gence of stories, their changes and paths of infection. Such networks

emerge when connections are made between social actors (“nodes”) –

which can be individuals or organizations. e collections of these con-

nections can be condensed into patterns or network structures that

describe the interaction of the system. It also provides a model for the

stories that remain in the resonant space of our echo chambers without

going viral. ese chambers, on the other hand, are not closed systems,

but rather incubators where new mutations emerge and then nd their

way out.

So if a virus now goes world-spirit, will virologists become the new

priests of our oracles? at’s exactly what Shiller said. And there’s a lot to

be said for giving them an important place on the interpretive team. After

all, to predict the spread of infection, we need to understand how people

behave collectively. To do that, you need the skills to understand how

infections are transmitted. But it also involves analyzing data and devel-

oping mathematical models. is is more of an interdisciplinary task, in

which the engineering sciences also have a lot to contribute, for example

when it comes to mathematical theories, in particular chaos theory and

statistical modelling. Together with other representatives from disciplines

such as mathematics and bioinformatics, they can develop the tools to

recognise patterns for which humans have no sense of their own and

therefore make use of the machine. And it needs the interpreter to place

these patterns in a larger context and make a coherent story from them.

If one conclusion can already be drawn from the Corona pandemic, it

is that data will play an even greater role in our daily lives in the future:

• e current crisis is giving digital communication a huge boost. is

means that much more data is being created in the industry’s systems,

especially by companies whose business model is based on this form of

data collection.

• Data is currently helping to improve our understanding of how the

virus is spreading. We have a real-time global pandemic development

laboratory that is providing us with an unprecedented volume of data.

H.-W. Eckert

• Data is the basis of the surveillance systems that will enable us to con-

trol, predict and manage our behaviour even better in the future.

Many of these practices already exist and are being used in the current

crisis to at least (and hopefully only) temporarily suspend fundamental

liberties in order to slow or prevent the spread of the virus.

us, the pandemic will further strengthen the role of data in under-

standing our world. Enormous sums of money are owing into their

collection and processing, new oracles are developing and with them new

priest boxes for interpreting all these new volumes of data, which are

constantly providing new material for interpretation.

In doing so, we simultaneously broaden and narrow what constitutes

our world. e philosopher Ludwig Wittgenstein once said, “e limits

of my language are the limits of my world.” e stories woven from lan-

guage make up this world, with its networks, nodes and echo chambers.

By means of data, we are able to advance into regions for which we

humans have no sense of our own. We see connections and patterns that

we would not have seen without this data. By evaluating and contextual-

izing this data through language, we expand our world. At the same time,

however, we run the risk of narrowing this world to that which exists

through data. We lose sight of the things that cannot be measured and

quantied. is is quite helpful to keep in mind when we ascribe so

much interpretive power to data.

References

Bruner J (1986) Actual minds, possible worlds. Harvard University

Press, Cambridge

Callaway E (2019) Is this cave painting humanity’s oldest story? Nature. https://

www.nature.com/articles/d41586- 019- 03826- 4. Accessed: 13. Dez. 2019

Campbell J (2008) e hero with a thousand faces, 3. Au. New World

Library, Novato

Duarte N (2019) Data story. Explain data and inspire action through story.

IdeaPress, Washington, DC

Dykes B (2020) Eective data-storytelling. Wiley, Hoboken

Fukuyama F (1992) e end of history and the last man. Free Press, NewYork

2 Storytelling in the Digital Age

Harari YN (2018) 21 lessons for the 21st century. Jonathan Cape, London

Hasson U (2016) is is your brain on communication, Youtube 03.06.2016.

https://www.youtube.com/watch?v=FDhlOovaGrI. Accessed: 25. Nov. 2019

Hobsbawn E, Ranger T (1983) e invention of tradition. Cambridge University

Press, NewYork

Horvitz J, Seetharaman D (2020) Facebook executives shut down eorts to

make the site less divisive. Wall Street Journal. https://www.wsj.com/articles/

facebook- knows- it- encourages- division- top- executives- nixed- solutions-

11590507499. Accessed: 1. Juni 2020

Kahneman D (2011) inking, fast and slow. Penguin, London

Kleine Wieskamp P (2019) Visual Storytelling im Business:– mit Bildern auf

den Punkt kommen. Hanser, München

Kuke U (2012) Wer schuf die Zeichen von El Castillo? Die Welt, 2. Ocktober.

https://www.welt.de/kultur/history/article108596340/Wer- schuf- die-

Zeichen- von- El- Castillo.html. Accessed: 25. Nov. 2019

Löer M (2014) ink content! Rheinwerk, Bonn

Lyotard J-F (1979) La Condition Postmoderne. Les Edition de Minuit, Paris

Pariser E (2011) e lter bubble. What the internet is hiding from you.

Penguin, London

Sammer P (2017) Storytelling. O’Reilly, Heidelberg

Seemann M (2018) Filterbubbles, ja, nein, doch, gerne, oder – Stand der

Debatte PIQD, 6. September. https://www.piqd.de/technologie- gesellschaft/

lterbubbles- ja- nein- doch- gerne- oder- stand- der- debatte. Accessed: 2. Mai 2020

Shiller R (2019) Narrative economics, how stories go viral and drive major eco-

nomic events. Princeton University Press, Princeton

Siefer W (2015) Wer erzählt, der überlebt Die Zeit, 23. Dezember. https://www.

zeit.de/2015/52/geschichten- maerchen- christentum- mythos- legende-

ueberleben. Accessed: 23. Jan. 2020

U.S. Army Heritage and Education Centre (2019) Who rst originated the

term VUCA (volatility, uncertainty, complexity and ambiguity)? USAHEC,

7. Mai. https://usawc.libanswers.com/faq/84869. Accessed: 29. Dez. 2019

Wason P, Johnson-Laird P (1968) inking and reasoning. Penguin,

Harmondsworth

H.-W. Eckert

From theQuestion totheData

Abstract Data sources ow so abundantly that when starting any project,

it’s important to ask precise questions. In this chapter, I show approaches

to using data to contextualize stories and help dene which audience I’m

targeting, which themes I’m setting, and why they’re relevant in the rst

place. e examples illustrate the challenges of developing the question

and in each case provide specic answers to key communication issues:

Why we tell something, who we tell it to, and what we tell.

To enter the land of milk and honey, the aspirant must rst eat his way

through a wall of dough. Once inside, he nds himself in a place where

food and drink abound. Wine ows from overturned jugs straight into

his mouth, pancakes fall from the sky, a roasted pig equipped with a knife

runs around as a mobile snack. is is how Pieter Brueghel the Elder

depicted the popular paradise in the sixteenth century.

e path to data sources is somewhat reminiscent of the farmer’s path

to the roast. An oversupply of food meets a limited stomach. In Breughel’s

painting, the peasant, like his fellow suerers, has capitulated and sur-

rendered to sleep. He is no longer even able to enjoy all the good things.

part of Springer Nature 2022

H.-W. Eckert, Storytelling with data, https://doi.org/10.1007/978-3-658-38555-2_3

To avoid a similar fate when processing data, we need a clear goal and a

lot of discipline to avoid succumbing to the obvious temptations.

ere are so many ways to tap into data sources that a precise question

is important at the start of every project. Only on this basis does it make

sense to develop your own data strategy for the project.

Companys usually oer a multitude of possible sources that can be

tapped. For communication topics, these are rst of all the core systems

in which customer data is stored. Ideally, it’s stored centrally in a cus-

tomer relationship system (CRM) and is also used and maintained by

several units of the company such as communications, sales and service.

In addition, all communication and marketing systems such as mailing

programs, websites, shop systems and apps are of course relevant for the

collection of data. So-called enterprise resource planning (ERP) systems

provide valuable sources, for example, about ows of goods and purchas-

ing behavior. In the meantime, more and more companies also have their

own so-called business intelligence, in which the data from the various

internal and external sources can be stored, analyzed and played back in

a central instance.

e advantage of self-collected data is obvious: You have direct insight

into its quality and signicance, know how it was collected and thus gain

access to important contextual information for its interpretation. And

you can access it at any time.

For most questions, however, companies are also dependent on exter-

nal sources. For communication and marketing topics, the channels on

which the company is active and in contact with its target groups are

ideal. Countless data sources can be tapped from media response analy-

ses, advertising impact studies and search engine evaluations as well as

from listening on social media channels, i.e. social listening. ey are

further supplemented by market media studies and other oerings from

market research rms. e interpretation of this data is usually more

complex because information about the method of collection, its quality

and signicance must rst be obtained and evaluated.

Regardless of where the data come from: ey are not themselves the

subject, but in this case they provide the narrative, i.e. the context of the

stories we tell (see Fig.3.1 and the comments on this in Sect. 1.4).

H.-W. Eckert

Fig. 3.1 Storytelling with data- the context of stories: In this chapter, data is not

itself the subject, but provides the context of the stories we tell. (Source: Own

representation)

In this chapter, I present selected examples of approaches that use data

to dene storytelling frameworks. ese examples show the challenges in

developing the question and each provide specic answers to the central

communication issues:

• Why we tell a story: e brand as an anchor of the narrative can be

better grasped in the perception of the target groups and the eect of

the communication measures.

• Who do we tell: the target groups, their needs, expectations and wishes

can be analysed.

• What we tell: We nd new stories in the data. And we discover ways to

connect them to the topics of our target groups.

3 From the Question to the Data

3.1 The Brand: Why AmIRelevant?

3.1.1 Adidas andtheQuantiﬁcation Bias

A failure of Google’s advertising server in Latin America in 2017 brought

Adidas surprising insights. After all, a large part of the advertising budget

there ran via Google AdWords and was intended to boost sales of sneak-

ers and sportswear. So it was natural to assume that when the servers went

down, so would sales. But what happened was: nothing. Although Adidas

could no longer place paid advertising, both trac and sales remained

constant.

e server failure brought Adidas a salutary insight: the model with

which the advertising budget had been distributed until then was wrong.

Far too much of it was allocated to so-called performance marketing, i.e.

primarily to paid search engine and banner advertising for specic prod-

ucts. e good performance and measurability of these advertising media

and the possibility of tracking them in the shop system had led Adidas to

place a special emphasis on this in its marketing. e server failure, how-

ever, made it clear that the company had overestimated this inuence.

Adidas global head of media Simon Peel used the lion’s den of his

guild, the annual performance marketers’ gathering in London, EFF

Week 2019, to admit the mistake: “We over-invested in digital advertis-

ing”. e focus on return on investment (ROI), he said, had led the

company to over-invest in performance-based advertising. Many of

Adidas’ products hit an oversaturated market where they were sold

through discounts, making customers increasingly price-sensitive. ere

was no long-term brand orientation. As a result, of the total budget, only

23% went to the brand, while performance marketing took 77% of the

advertising budget.

is budget was directly tied to e-commerce sales. Behind this was the

belief that digital ads drive sales. Adidas was keen to drive online sales, as

this was the most protable part of the business. “We assumed it was

digital advertising– desktop and mobile– that was driving those sales,

and as a result we were over-investing in that area,” Peel said. e insight

led Adidas to introduce an econometric model that brought together

brand, sales and customer structure. Until then, the company had

H.-W. Eckert

believed that loyal customers drove sales. But the results from this model

revealed that 60% of revenue came from rst-time buyers.

Adidas also had to abandon another basic assumption: Namely, the

belief that business units only drove their own sales, so soccer advertising,

for example, would drive sales of soccer boots. In reality, all advertising

paid into the sale of Adidas products, i.e. the soccer boot also paid into

the jogging outt.

After all, Adidas’ old model was based on tunnel vision that looked at

each channel in isolation. Little was invested in video because it didn’t

perform particularly well in last-click metrics. But a closer look showed

that TV, out-of-home and cinema were also paying into e-commerce.

e new model highlighted that brand activity was responsible for 65%

of wholesale, retail and e-commerce sales, while performance marketing

conversely also drove wholesale and retail sales.

So it was time to shift the allocation of the advertising budget signi-

cantly towards the brand. Adidas used the four years following the server

outage to develop a new model and approach for it. With its marketing

playbook– called ‘Creating the New’– Adidas introduced a new cam-

paign framework in 2019 that was focused on emotional brand activa-

tion. ree to four major campaigns per year are expected to recharge the

brand in the future (Peel 2019).

is shift from marketing eciency to marketing eectiveness made

waves. e industry discussed little else in the aftermath. Peel’s speech

was a kind of wake-up call, in the wake of which other advertisers

announced that they would rethink their performance share of their mar-

keting mix, including Booking Holdings (Booking.com, Priceline,

Kayak), Tripadvisor or the Old Navy brand from the Gap Group

(Rentz 2019).

Adidas is a good example of the quantication bias that many compa-

nies succumb to. It’s just too tempting to surrender to tunnel vision and

focus on digital advertising eorts and here especially the last click. is

data is readily available in large numbers. It can be easily tracked, seg-

mented and analysed, making it easy for the communications manager to

make a numbers-driven controller and perhaps many a CFO happy by

explaining to them how to derive impact relationships, ROIs and the

budgets required. It is more dicult to get a view of long-term

3 From the Question to the Data

developments of a brand and cross-channel eects. Such a model is costly

to develop and needs a common understanding of the entire manage-

ment on the impact correlations. As a rule, the gures for this have to be

collected rst.

But it’s worth taking this route and not falling into the quantication

trap. By investing their brand(s), companies build a web of narratives that

helps them stabilize their sales (see Adidas) and also builds them a buer

for times of crisis. General Electric, for example, lost around 60% of its

share value during its downturn from 2017 to 2019, but only 40% of its

brand value (Economist 2020). e fact that it can also go the other way

and a decision can destroy brand capital that has been painstakingly

acquired over decades is again shown by the example of Adidas. With the

decision not to pay rents for the stores during the Corona crisis, Kasper

Rorsted incurred the wrath of many loyal fans of the brand. Even though

Adidas reversed this decision after the storm of indignation: the damage

was there and can probably only be repaired in the very long term. Adidas

is now at the forefront of companies that only look out for the good of

shareholders and shirk their responsibility to society, especially in times

that call for solidarity. is is not a good story for a brand that wants to

connect people through sport.

3.1.2 Volkswagen: Brand Communication

withBig Data

e question of why a brand exists can only be answered by the company

itself. e power it needs for this positioning comes from the culture, the

shared experience and the ideas about future development. On the other

hand, to balance the internal and external views, it needs sound market

research. Amazon founder Je Bezos summed it up as follows: “A brand

is what people say about you when you’re not in the room.” Only with a

clear picture of how a brand is perceived by its target groups can it be

determined what steps are needed to bring the target closer to the actual.

So in Bezos’ words, matching what people say about me when I’m not in

the room with how I want to be seen as a company. at forms the basis

of all brand management.

H.-W. Eckert

Data is essential when it comes to the perception of a brand. e

example of Volkswagen shows how big data can support this. In the time

before the diesel scandal, brand perception at Volkswagen was character-

ized by values such as discipline, precision and reliability. A sober, objec-

tive, one could say engineer-driven positioning with a penchant for

clearance. e brand’s reputation has suered badly as a result of the

diesel scandal. Its credibility was heavily aected after the interventions

in the exhaust gas purication systems. In order to reposition itself among

the relevant target groups, Volkswagen has taken an unusual route.

e company began with a new accentuation of the core values. Out

of the eld of factual values around the concept of discipline and into the

eld of emotional values around the topics of experience and joy. Here,

the focus was now on virtues of being innovative, valuable, responsible,

credible, fair and reliable. In order to enliven them with a new story,

Volkswagen specically sought insights into the expectations and atti-

tudes of the target groups. is is usually done by interviewing focus

groups and online panels. Here, Volkswagen gave the country managers

creative freedom. In Poland, for example, Volkswagen relied on the “Flash

AI” Big Data platform from NeuroFlash. is is a kind of semantic data-

base that makes it possible to discover and weight associations of words.

In this way, attitudes of target groups can be determined without inter-

viewing them. To this end, NeuroFlash has compiled articles from online

sources such as Wikipedia, mass media and social media channels and

evaluated them using machine learning methods to identify and weight

word meanings and associations.

Volkswagen used the platform to identify characters that t the brand

particularly well. Here, two top occupational groups were found through

semantic analysis: the doctor and the teacher. As an intersection of these

two characters, the two core ideas from which the campaign ideas were

developed emerged: “schoolchildren” and “rst aid”. e “Little Heroes”

campaign focused on teaching rst aid to school children and their fami-

lies (see Fig.3.2).

e result: a more than fourfold higher interaction rate compared to

the successful T-Roc/Tuareg campaigns on Facebook. Brand-relevant

indicators such as brand consideration were also clearly addressed. In

addition, the campaign still paid into Volkswagen’s CSR perception

(Mall 2020).

3 From the Question to the Data

Fig. 3.2 Using a semantic database, characters were identiﬁed for VW that ﬁt

the brand particularly well. From this, the two core ideas crystallized, from which

the campaigns were developed

e example shows: When developing a brand message, the internal

view is not a good guide. It is important to translate the “Why?” into the

language of the target groups and to connect with their narratives. New

methods such as the semantic analysis of words and the identication of

association elds on the basis of big data provide important impulses

here, can be implemented faster and with less eort than classic market

research and will continue to gain relevance.

3.2 The Target Group: ForWhom

AmIRelevant?

KYC is not a new fast food chain, but stands for Know Your Customer.

In the data age, knowledge about the wishes, needs and situation of cus-

tomers is increasingly becoming the most important success factor for

companies. is is because we have very powerful sources through digital

channels that provide valuable insights into customer attitudes, expecta-

tions and desires. ose who use them correctly are ahead of the game.

is can be done in very dierent ways. Bottom-up or top-down, based

on available data or based on a strategic decision. But best of all based on

a combination of both.

H.-W. Eckert

3.2.1 An Online Shop Sharpens Its Customer Proﬁle

e shoe retailer Seven Feet Apart from England has chosen a bottom-up

approach to optimize its target group appeal (www.sevenfeetapart.com).

e company sells high-quality shoes exclusively through its online shop.

Two years after its launch in 2016, the company had collected a lot of

data that was suitable for analyzing and clustering its customers and

allowed it to review and develop its original premises. e company used

data from website analytics, in this case Google Analytics, data from its

Shopify storefront system, and data from its Dotmailer email program.

is was supplemented by customer demographic data and– very impor-

tant for the mail order business– return rates. On this basis, the company

developed a data platform that helped it understand and segment its cus-

tomers and oer them the right shoes at the right time.

Using the clusters of Experian, a leading market research company

from England, Seven Feet Apart segmented its own customer groups. In

this way, customer personas could be developed via the Experian tool.

Personas are personalized, typical representatives of a particular segment.

ey can be used to gain deeper insights into attitudes, expectations,

desires, values and, in particular, purchasing behavior.

With the help of the personas, the company was then able to model

the target customers as a projection of the existing customers. e basis

for this was the identication of the most valuable customers, which were

determined on the basis of purchase volume, frequency and the time of

the last purchase. To expand this customer group, Seven Feet Apart

focused on nding and targeting statistical twins. is was done by ana-

lyzing the engagement of the corresponding customers on the channels

and their preferred devices. is allowed the questions to be answered:

• How do most valuable customers dier from everyone else?

• How are the most valuable customers acquired and retained?

• How do these customers shop?

• What triggers a purchase?

3 From the Question to the Data

Among other things, the company found out that the age of this core

target group is not between 35 and 45, but between 45 and 60. is led

to some changes in the approach: the selection of models in the cam-

paigns, the product range and its presentation as well as a stronger focus

on values in general and on sustainability in particular. For example, the

product presentation of the shoes was supplemented by story elements,

i.e. alternative entrances combined via colours, themes or occasions.

With the 100 Square Feet project, the company also addresses the sus-

tainability issue: with every purchase of a shoe, Seven Feet Apart donates

to the World Land Trust in order to preserve 100 square feet of rainforest.

Another important insight was that the most valuable customers were

also the wealthiest. is led the company to evaluate its pricing strategy.

e fact that price had no great impact on conversion rates and that cus-

tomers were willing to accept higher prices conrmed this move.

e analysis and classication of customers has brought more prot-

ability to Seven Feet Apart. Above all, it created the prerequisites to prop-

erly address this protable customer group, to select the appropriate

products and to set the right topics (Aaron 2019).

Communication and marketing units always have a potential hero, the

customer in all his facets. e persona concept provides an important

approach to identifying these customers, their needs and desires. For the

development of stories on the basis of personas, data is the foundation: It

ows in from the various contact points– the purchase history, the visit

to the website, the opening behavior of newsletters, the contacts with

sales and service employees, some of them possibly bundled in a CRM or

BI system. In this way, it is possible to address age groups with particu-

larly high purchasing power, their topics and wishes. ese can be supple-

mented by external sources that provide information about the interests

and relevant topics of these personas.

On this basis, the dramaturgy can then be developed for each indi-

vidual persona. e journey of this persona, the customer journey, is a

narrative concept that is fed by data. is dramaturgy can be orchestrated

along the so-called touchpoints. Data helps to understand at which point

this persona is and which impulse he needs next. In fact, he should ana-

lyze exactly which typical journeys there are for his oer and at which

points he loses particularly many travelers.

H.-W. Eckert

Such customer journey analyses quickly become very complex. Here, a

core competence of our brain can be used: We are masters of pattern

recognition. And the fastest way to recognize patterns is through our

sense of sight. is helps us to analyze complex relationships. For exam-

ple, in the question of what the customer’s journey to the company looks

like. Such an analysis brings together an enormous amount of data –

from a wide variety of sources, such as ad tracking, the website, social

media channels, the shop system, the email system. By visualizing this

data, it is possible to see where most customers come from, through

which channels they are reached and which path they take to purchase

and beyond. Especially in such complex contexts, humans are sometimes

superior to machines.

3.2.2 A Fashion Retailer Reinvents Itself

If there is more than one distribution channel, it becomes much more

complex. Multichannel is the term retailers use to describe their claim to

meet customers on the exact channel they are using. Whether they are

browsing through the catalog or scrolling through their Instagram time-

line, it is important to set exactly the right impulse to lead them in the

right direction on their journey until they nally place a product from

the retailer in their shopping cart and order it. However, the change from

a goods economy to customer-oriented thinking also requires the consis-

tent realignment of all processes in the company, a good data strategy on

the basis of which the paths of the customer can be tracked.

Multichannel is a lofty claim that is best summed up this way: e

easier the purchase is for the customer, the more complex the sale is for

the company. After all, it’s not just about managing the many channels so

that they address the right person at the right time with the message

tailored to them. In most cases, the data behind them is in dierent silos:

A good example of this is the analysis of the customer journey of applicants of a personnel service

provider. With the visualization methods of the process mining tools, it was not possible to under-

stand the behavior patterns of the applicants. is was nally achieved by visualizing detailed

processes. On this basis, not only the data specialists, but also many employees from the most

diverse departments were then able to analyze the processes and created the basis for a broad and

fruitful discussion about optimization potentials: Münster (o. J).

3 From the Question to the Data

knowledge about the goods (usually in the ERP systems) and knowledge

about the customers (usually in the CRM systems) are not networked

with each other; moreover, data from online and oine channels is often

held in dierent systems. A change of the customer from one medium to

another can currently still be tracked well with cookies on online chan-

nels. But how do I bring together the activities from the online shop with

the retail stores or the catalogue? How do I manage to bring together a

uniform view of goods and customers in such a way that I can draw the

right conclusions from them and give appropriate impulses in real time?

It is clear that retail companies, as soon as they have more than one

distribution channel, face particular challenges here due to the abun-

dance of data and systems involved. e example of fashion retailer

Atelier Goldner Schnitt shows how the right course can be set. For the

longest time, the company had only one sales channel: at was a high-

quality catalog of fabric samples that customers had to return with their

order in a specially designed envelope. e company’s brand promise:

fashion in a perfect t for an older target group. e average customer

was 83years old, and the business model was very stable for a long time.

But over time, the expectations and experiences of this target group

also changed, and new competitors came along who used digital sales

channels and put pressure on earnings. is made a reorientation

necessary.

is reorientation was essentially based on a rejuvenation of the target

group and the introduction of two sharply dened customer proles. In

addition to the traditional regular customer, a further, somewhat younger

customer prole was developed. In addition to the age of the target cus-

tomers (83 vs. 68years), these proles also dier in terms of their fashion

tastes, their online anity (9% vs. 50% online use) and the cancellation

rate (low vs. higher).

ese customer proles were not only mapped in the brand world, but

in the entire organization– from purchasing to the assortment to the

customer approach. e new customer group also brought other contact

points into play: above all the web shop and e-mail communication. An

important point of the entire reorientation was the data organization.

H.-W. Eckert

is was realized by a data model centered on the customer: All contacts

online and oine are stored with the customer and thus provide a holis-

tic view of the relationship. is is a great challenge, especially for com-

panies from the stationary trade. is is because the web shop is usually

developed at some point as a second channel alongside the stationary

activities. With a separate customer and inventory management and a

completely separate approach, you then have two silos and noth-

ing gained.

Atelier Goldner Schnitt avoided this mistake by focusing on a strict

customer centricity from the very beginning of the reorientation. e

CRM system quickly reached its limits. e challenge was to map the

increased complexity and to draw the right conclusions from it as auto-

matically as possible: to integrate all information from the contact points,

including outbound measures such as calls, mailings, newsletters, and

orders from the catalog and the shop, and to derive customer loyalty

activities and impulses for customer development (rst to second pur-

chase) from this.

is was realized through an interaction of a so-called scoring engine

(Gpredictive), a marketing automation tool (Cross-Engage) and a busi-

ness intelligence platform (incuda). e BI platform provides the aggre-

gated data about customers and products, the scoring engine calculates

with the use of machine learning continuously and automatically what

the future sales of the customers look like. e marketing automation

solution in turn controls the corresponding actions in the appropriate

channels.

e basic decision was of utmost importance: the entire data budget

was aligned with the female customers. With the introduction of two

distinct proles, the company is able to gather valuable information

about the behavioral patterns of its female customers. is means that the

company is now in a position to develop a precise approach for the two

groups of female customers– across all channels. On this basis, two fash-

ion worlds could be created and transported on the corresponding com-

munication channels (Anton 2019).

3 From the Question to the Data

3.3 The Topics: What Is Attracting?

3.3.1 What Advertising Does theCustomer

Want toSee?

John Wanamaker, the so-called department store king and father of mod-

ern advertising, is reported to have once said, “I know half my advertising

is money thrown out. I just don’t know which half.” Since then, an entire

discipline has been devoted to the study of these two halves: so-called

advertising pretesting. It aims to improve the ratio in favor of eective

advertising. Because ad space costs a lot of money. And it’s better to invest

some of it in testing procedures so that you can then advertise more eec-

tively. As a rule, this involves testing an advertisement or campaign in a

kind of laboratory situation before it is placed, to determine whether it

can meet the requirements placed on it or whether and in what form it

can be optimised (for more on this, see Trommsdor and Becker 2009).

But in the meantime, there are also automated processes that deter-

mine the probability of success of an advertising spot on the basis of

specially developed metrics. e soft drinks giant Coca-Cola, for exam-

ple, increases advertising eectiveness using its own “One Number

Score”. is is a metric that can be used to indicate whether creations are

achieving strong business results and appealing to consumers. Spots have

to land in the top quarter of market researcher Kantar’s creativity data-

base to ultimately make it to market.

is was originally a big shock to the system– to the ad agencies and

the marketers who weren’t used to testing. It was also important to edu-

cate the internal brand managers and their agencies about how the pre-

testing system works and to get them involved so that they have an

indicator of how the spot might work before production.

Now, however, Coca-Cola is also automatically assessing the likelihood

of success of new spots on the basis of a large database with over 30,000

videos. e technology relies on facial recognition methods and uses neu-

roscientic approaches to screen out the best spots. In this way, the com-

pany shifts pretesting to the machine and makes the results from one

country transferable to other countries (see Kantar 2020).

H.-W. Eckert

Other methods are also being developed here. For example, the com-

pany Aiconix AI wants to recognize at which point a viewer of a lm

breaks o and automatically cuts the lms together so that the viewer

continues watching. is type of optimization should lead to lms hav-

ing a higher acceptance and eectiveness on social media channels.

3.3.2 Listening toWhat Moves theUser

Where the most questions are asked, that’s where the most knowledge

arises. e priestesses of the Oracle of Delphi already knew that. Success

fueled the business model, which at some point proted from its own

fame and provided the best answers thanks to the many questions asked

from all sides. Today’s search giants, led by Google and Amazon, are no

dierent. While Google “only” monetizes knowledge in ads, Amazon also

has the opportunity to let its users generate ideas for new products and

add them to its portfolio. So the platforms themselves benet the most.

But companies can also use this knowledge via various tools to keep an

eye on relevant topics and to know what the audience wants to read.

Even in the more consulting-intensive B2B business, digital channels

such as Internet searches and videos account for almost 60% of informa-

tion procurement. is was revealed by a survey of 2745 sales managers

conducted by Google and the consulting rm Roland Berger in 2015

(Roland Berger 2015, p.6). Given the growth of digital channels, it can

be assumed that this gure has, if anything, risen even further in recent

years. Here, it is a matter of attracting attention in the early phases of the

customer journey in order to become relevant to the potential buyer in

the rst place.

Since in the early stages there is usually still little knowledge about a

concrete solution or product, search movements are more focused on

advice, problem solving, product comparisons and evaluations.

Take the purchase of a mattress, for example. is happens relatively

rarely in most households. Hardly anyone will therefore type the name of

a manufacturer into the search engine line at the beginning of their

e use case is still in development, talk Eugen Groß, AI in Marketing, IHK Munich 28.01.2020

and https://www.aiconix.ai/anwendungen/

3 From the Question to the Data

search. A strategy based on a strong brand will not be relevant here for the

acquisition of new customers, but rather for the retention of existing

customers and their activation as referrers. Any provider looking to

expand their reach beyond their existing visitor and customer base will

therefore be on the lookout for topics that may be relevant to a potential

customer. e aim here will be to provide answers to more general ques-

tions, such as: How low in toxins is a mattress? How can a mattress be

cleaned? How often should it be turned? Do I need two mattresses for a

double bed? But a mattress can also solve a specic problem: e.g. How

does the mattress relieve my back pain? How does it support a peace-

ful sleep?

Search queries provide an insight into what moves the potential cus-

tomer. In order to approach the right topics here, the data that users enter

into the search engines helps. Many tools have emerged around the major

search engine operators that help to understand the interests and lan-

guage of users. Keyword databases help nd the right search terms– they

provide related terms and longer phrases. ey can be used to discover

trends and cycles. ey provide information on how certain topics

develop over time– whether they uctuate seasonally, decrease or increase.

Although search volumes are often highly rounded and search queries

never provide a complete picture of all possible queries, they are a valu-

able and powerful tool for identifying topics and trends.

In addition to the potential customer, you also need to look at the

competition. Which competitor is doing a particularly good job? is

can be seen from two things:

• Backlinks

• SEO relevance

Backlinks: Indicate how many links point to your own website or that of

a competitor. ey provide an insight into the quality and relevance of

the competition and of course also information about which pages are

worth linking to. Link databases provide the material to analyze the link

strategies of competitors. e providers of these databases use them to

build their own web index. is has a certain subjectivity, which results

from two things: On the one hand, the link data are not always complete

H.-W. Eckert

and do not provide a complete or up-to-date picture of the situation. On

the other hand, they are only ever approximations of the search engine

operators’ algorithms. Nobody knows how Bing and Google really count.

SEO relevance: Once your own topics and keywords have been iden-

tied, they should also be analysed in comparison to your competitors.

e tools provide an important indication of how visible one’s own web-

site is compared to competitors. Specically, they answer the questions:

• Where does a competitor rank in front?

• How do these positions per keyword develop over time?

is allows conclusions to be drawn about content strategies and also the

“quality” of content (from the point of view of the search engines and

their algorithms, not from the point of view of the people who read the

content!) Even if the tools do not capture all rankings and are rather dif-

cult in niche markets, they provide important indicators for the visibil-

ity of one’s own content strategy in the competitive environment.

If you have identied the relevant backlinks and developed your own

strategy, where your own backlinks are desired, a link building oers

itself. is is an important part of the so-called content seeding. e aim

of the procedure is to intercept potential customers where they already

are by spreading links to one’s own content. Tools help to target websites

for specic topics that want to link to or mention a piece of content.

Here, there are a number of tools that help identify and categorize rele-

vant search terms (see Fig.3.3).

Which tool is the right one depends above all on the specic needs of

the user– and of course also on the price he is willing to pay. ere is a

whole range here– for the beginner and for the professional, for the web-

site of a start-up and that of a corporation, free of charge or with a three-

digit monthly fee: free tools such as Seobility, Seorch and Ubersuggest

have a limited range of functions, others are specialized in certain topics:

Depending on whether keyword rankings, backlink or competitor

research, the choice will fall on one of the tools. Since the keyword indi-

ces are always based on the quality of the provider data, they also dier

according to topic, industry and region.

3 From the Question to the Data

Fig. 3.3 Search queries provide an insight into the interests of potential custom-

ers. The search engine operators’ data can be evaluated according to different

criteria using various tools. Interactive graphic at www.data- storyteller.de.

(Source: Own representation)

Almost all of them make it possible to monitor competitors and thus

determine at any time who is ahead with which search term. is pro-

vides valuable information about where there is room for improvement

in the search term strategy.

In addition, some tools also oer semantic methods that can be used

to identify not only search terms, but entire topic areas. In this way, con-

tent managers receive important information about which topics interact

with other topics and where synergies may arise.

Of course, not only search queries are decisive in determining the top-

ics of the target groups. Everything that is written, lmed, commented

on and redistributed plays a role here. Television, radio, print media, the

Internet and social media form a huge pool of sources for these types of

analysis. Here, press and blog articles, radio, TV broadcasts and video

contributions, social media posts, likes, comments and shared content

are included in the evaluation.

is gives communication and marketing managers valuable ideas for

discovering trends, identifying media representatives and inuencers, and

nding suitable channels for disseminating their own topics (see Fig.3.4).

H.-W. Eckert

Fig. 3.4 Content marketing tools can be used to discover trends, identify inﬂuencers and ﬁnd suitable channels for dis-

seminating one’s own topics. Interactive graphic at www.data- storyteller.de. (Source: Own representation)

3 From the Question to the Data

Among other things, the tools help to discover topic areas, identify

channels and nd multipliers such as bloggers, inuencers and press rep-

resentatives through whom content can be further disseminated. ey

make it possible to compare oneself with competitors, to gauge the mood

for topics, to plan one’s own content and to distribute it on social media

channels. Many also have integrated monitoring, which can be used to

monitor key gures such as click rates, engagement and follower growth.

ere are major dierences in the depth and breadth with which the

sources are integrated. As the overview shows, there is a focus on social

media in all tools. Other channels such as print and radio/TV are only

integrated by a few providers. is is also a cost issue, as the preparation

of these sources is costly. Here, companies like Argus, Echobot and

Unicepta, which come from media resonance analysis, are ahead. Google’s

Ngram Viewer occupies a special position in the list. It covers a much

longer period than all the other tools and is based on the so-called Ngram

Corpus, digitized books from ve centuries up to the year 2019.

In general, the same applies to all tools: It is worth taking a very close

look and asking exactly which sources are tapped in the categories men-

tioned. Here, the tools dier considerably. Also, not all social media is the

same: ere are dierences in the channels monitored. And within the

channels, the range is enormous. No tool oers comprehensive access.

is is not even technically possible, as the interfaces do not allow this.

So only selected, publicly accessible posts can be extracted. In order to be

able to evaluate these, the providers have to provide appropriate storage.

Storage space is cheap, but the amount of data is enormous. erefore, all

providers only have selected content over a certain period of time. What

these are depends primarily on how much money a provider puts into the

storage solutions and who the main users of the system are. So the users’

willingness to pay and their areas of interest have a big impact on the

sources available. Companies should take a close look at which focal

points are stored there and what options there are for incorporating indi-

vidual requirements.

Here, too, the following applies: ere is a wide range and it is worth-

while to think about the specic use case in advance before you start

looking for the right tool. Because it makes a dierence whether I want

to address a B2B or B2C market, search for the right inuencer or

H.-W. Eckert

primarily produce content myself, which channels I use to address my

customers, how I plan, produce and play out my content and according

to which key gures my communication is controlled and the success is

measured.

3.3.3 A Viral History ofArtiﬁcial Intelligence

When it comes to processing data, methods from the eld of articial

intelligence (AI) are being used more and more frequently. It has long

been impossible to imagine our everyday life without these technologies.

e recognition of language and images, for example, is not possible

without the use of AI.But the methods are still used in many other appli-

cations and are currently being tested in new elds. In this book, we

encounter these technologies in several places and see examples of how AI

supports the discovery, telling, and interpretation of stories.

At this point, however, I will show how the topic area around AI itself

has developed as a narrative and gone viral. For this purpose, Google

provides the Ngram Viewer and Trends, two tools for a practical and

freely accessible evaluation. With Trends, Google presents an aggregated

overview of its search queries. ere, various search terms and topic areas

can be entered, combined and displayed in a graphical progression. e

tool thus provides valuable indicators for discovering relevant trends– for

the period from 2004 to today.

A much longer period is covered by the Ngram Viewer, which also

oers a visualization of keywords. ese are based on the so-called Ngram

Corpus, which are digitized books from ve centuries in eight languages,

representing 6% of all books ever published, according to Google (Lin

etal. 2012). e time frame ranges from the year 1500 to 2019, making

it a good tool to observe longer trends. Unlike search queries, mentions

on Ngram show the result of engagement with the topics covered in the

books. ey are thus slightly later indicators than a search query, but that

doesn’t matter as much in the long run. Over a 15-year period– between

2004, when trends began to be recorded, and 2019, when Ngram’s record

ends– the two tools overlap. Unfortunately, you look in vain for absolute

numbers from both Ngram and Trends. ey only provide percentage

3 From the Question to the Data

values, so that no statement can be made about how large the actual

amount of keywords or search terms is.

In order to enter the matching terms, Ngram requires some prior

knowledge. us, it is useful to identify word elds, associations and

related terms in order to nd matching tracks. is can also be done

using tools that evaluate association elds and semantics. However, none

of them has a data history going back to the previous century. In this case,

we would miss out on the term “cybernetics” (CY), for example, with

which the US mathematician Norbert Wiener laid the foundations of AI

in the 1940s.

If you search the English corpus of Ngram for the terms “Articial

Intelligence” (AI), “Cybernetics” (CY), “Machine Learning” (ML) and

“Neural Network” (NN), you will see the picture in Fig.3.5.

We see here

four dierent progressions of an infection, to speak with Shiller. ese

four progressions can be used to tell the story of the development of arti-

cial intelligence.

A rst, at curve is formed by the term CY, much steeper curves by the

two terms AI with a peak in 1989 and NN, the latter oset by about ten

years in 1997. ML, on the other hand, slowly but steadily establishes

itself from about the 1980s in the slipstream of the two steep curves and

overtakes the term CY in terms of frequency of mentions at the begin-

ning of the new millennium.

e term CY goes back– as already mentioned– to Norbert Wiener.

Since 1943, the US-American mathematician– spurred on by the entry

of the USA into the war– thought about how the behaviour of ghter

pilots could be predicted in order to be able to shoot them down better.

His 1948 work “Cybernetics– Or Control and Communication in the

Animal and the Machine” laid many of the foundations of AI. Most

importantly, he formulated the idea that there was no essential dierence

between humans and machines. is is a breakthrough for the develop-

ment of the later idea that the human brain is nothing but “wetware” and

thus follows the same rules as a computer with hardware and software. In

his words, “In fact, the whole mechanist-vitalist controversy has been

“Big Data” doesn’t yield relevant results in Ngram until 2007, while “Algorithm” yields too many

that have nothing to do with AI.

H.-W. Eckert

Fig. 3.5 The ﬁeld of artiﬁcial intelligence is developing in several waves. Since

2011, it has started another hype cycle, Interactive graphics at www.data-

storyteller.de. (Source: Google Ngram Corpus English 2019)

relegated to the limbo of badly posed questions” (Wiener 1948, p.44).

But if the human brain and a computer function in the same way, the

precondition is also created for seeing man and machine in a fruitful and,

in some eyes, fearful synthesis, as reected, for example, in the ideas of

transhumanism (see, for example, Lordick 2016).

rough the musician John Cage, Wiener came to John Brockman,

the initiator of the Serpentine Marathons in London, a dinner series in

3 From the Question to the Data

which he gathered the leading minds of his time (Kreye 2018, p.23).

Today he would be called an inuencer. Brockman made Wiener and his

ideas famous, as can be seen in the curve that peaked after Wiener’s death

in 1964.

e term AI rst appeared in the mid-1950s. It was invented by John

McCarthy, a US logician. At the time, he needed a powerful term for a

grant application to the Rockefeller Foundation for a conference in

Dartmouth the following year (McCarthy et al. 1955). Wiener had

already laid the groundwork for the man-machine narrative, but the

neologism AI condensed this narrative into a much more powerful term

compared to CY.e concept won over the Rockefeller Foundation, and

the grant application was approved. us, in the summer of 1956,

Dartmouth College became the birthplace of AI.e term took o for an

initial success story in the 1970s, then nally took o steeply in the

1980s. Wiener was not invited, his cybernetic concept was ignored by the

initiators of the conference.

In the years from 1970 to 1975, public and private funding for AI

research declined, investments were cut and start-ups in the eld received

less support. is phase is referred to as the rst so-called AI winter.

Strong, exaggerated expectations led to a “trough of disillusionment” of

the hype cycle in those years (see on Gartner and the hype cycle Sect.

1.2). Interestingly, the rst so-called AI winter only led to a attening of

the CY curve, while the term AI continued to be popular in the litera-

ture– even if it had not yet reached the level of CY.Research and publica-

tions on the topic of AI, on the other hand, continued to grow.

en, in the mid-1970s, a practical use returned to research with

expert systems and also gave the term AI a further boost. AI then reached

its rst peak towards the end of the 1980s. After expectations of expert

systems also proved to be overblown, a second AI winter dawned in 1987,

which lasted into the nineties.

A new concept gained the attention of researchers in these years: the

concept of neural networks, which had already been developed in the

1940s, was decisively advanced by the Japanese computer scientist

Kunihiko Fukushima: in 1975, the concept of the cognitron emerged,

which Fukushima expanded in 1980 as the neocognitron, thus laying the

foundations for a so-called Deep Convolutional Neural Network, which

H.-W. Eckert

was used for the recognition of handwriting and other visual patterns

(Fukushima and Miyake 1982). With this concept and through the tech-

nological development in the 1980s, neural networks got the decisive

push. For the rst time, it was possible to implement nodes and networks

based on the structures of the human brain using powerful computers.

is goes hand in hand with Marvin Minsky’s concept of distributed

systems, which he formulated in his 1986 book “Society of Mind” and

which describes how simple building blocks can be used to solve complex

problems through interactions. Minsky, incidentally, was a co-initiator of

Dartmouth College with McCarthy. Even though powerful computers

became available in the mid-1980s, raising hopes for further develop-

ment of neural networks: once again, data volumes and computing power

were not enough to satisfy expectations in the technology. e second AI

winter fell in 1987, this time clearly evident in the decline of AI’s uses (A

good overview of the history of AI is provided by Manhart 2018). But

this winter also set the stage for the growth of something new: the con-

cept of the NN took o here, then replaced the notion of AI in the hit list

in the early 1990s, peaking in the last third of the 1990s.

As a subeld of Articial Intelligence, the term ML takes on a very

dierent trajectory. e rise has been slow but steady since the mid- 1950s.

In the early 1960s, it was roughly on par with the term NN, but the latter

was subsequently used much more frequently. e rise of ML, on the

other hand, continued to be steady but slow until the late 1970s, when

mentions of it also began to rise signicantly. ML is based on research in

pattern recognition, which uses mathematical and statistical models to

learn from data sets. Neural networks form the basis of ML, in that the

success of one term pays into the rise of the other.

e development of AI remained a topic for experts for a long time.

But shortly before the turn of the millennium, events staged at great

expense were to give the subject area more resonance among the public.

IBM kicked things o with the series of chess matches between the chess

computer Deep Blue and the world champion Garry Kasparov. Kasparov

won the rst match against the chess computer in Philadelphia in 1996.

In the second match in NewYork in 1997, Deep Blue was victorious.

is game was the rst victory of a machine over a world chess champion.

e documentary lm e Man vs. e Machine picked up on the theme.

3 From the Question to the Data

e victory divided experts: some described it as a milestone in AI

research, others as a dead end, since the chess computer based its superi-

ority on pure computational power, which had nothing to do with real AI

(Heßler 2017).

Even though the competition, with its narrative of man’s struggle

against the machine, generated a great deal of media coverage, it did not

pay o in terms of further publications on the subject area. Publications

on AI had been in decline since 1988, those on NN had peaked in 1995,

and even for ML there was no impetus from the event. For search queries,

unfortunately, there is no data on this event yet, even if Google was

already online.

IBM itself did not associate Deep Blue with machine learning meth-

ods. e company’s communication was aimed more at supercomputing,

i.e. powerful mainframes that could rival the human brain with sheer

computing power. Deep Blue’s success was based on the fact that it made

its decisions of analyzing several thousand games. With its high comput-

ing power, it used the analysis to calculate the next moves. e method is

also called Brut Force. It is used when there are no known ecient algo-

rithms that can solve the problem. e most natural and simplest

approach to an algorithmic solution to a problem in this case is to try all

potential solutions until the right one is found.

Sympathies in this battle

between man and machine were clearly on Kasparov’s side. His support-

ers consoled themselves with the fact that at least the computer couldn’t

gloat over his victory. e supercomputer, even with its brute force

approach, played more the role of a muscular but cold and somewhat

simple-minded Goliath.

IBM made another attempt to reach a wider public in 2011. Under

the name Watson, the company presented a computer program from the

eld of articial intelligence. e program was developed as part of the

DeepQA research project and was able to provide answers to questions

that people asked it in natural language. Watson used speech recognition,

a sub-discipline of machine learning, to do this. To prove its capabilities,

Watson competed on the quiz show Jeopardy in February 2011 against

two human opponents who had previously won record amounts of

But even on the term “supercomputer” the action did not pay o, as an evaluation on Ngram shows.

H.-W. Eckert

money. A prize of one million dollars was oered for the match. e

media picked up on the theme, comparing the contest to Deep Blue’s

duel with Garry Kasparov. Again, the big theme was the battle of man

versus machine. After human Jeopardy opponents Ken Jennings and Brat

Rutter were tied after the rst round, Watson emerged as the clear winner

in the following two rounds. IBM focused on the human versus machine

theme in both competitions. In communication, this antagonism is dan-

gerous: the sympathies are with the humans in this case. e machine is

possibly admired as a dangerous opponent, perhaps even the technical

engineering achievement is appreciated. What is decisive here, however,

are the ideas of struggle, opposition and technical superiority of the

machine that defeats man.

e success of the spectacular Go game series of Google’s AlphaGo

machine learning system may have played a signicant role in the grow-

ing public interest. In October 2015, the computer had already defeated

a professional gamer. In March 2016, the AlphaGo team nally managed

to beat the reigning Go world champion Lee Sedol in the fth round on

move 37, winning a total of 4 games.

Google has not staged the communication of the topic as a battle of

man against machine, but as a sporting competition between a program-

ming team and an ingenious Go player. ere are people on both sides–

that is the crucial dierence in communication, which the lm about the

event “AlphaGo – e Movie” (https://www.youtube.com/watch?v=

WXuK6gekU1Y) also highlights. is gives the AlphaGo team a chance

to gain sympathy. And the decisive contribution is made by Lee Sedol

himself by talking about the beauty of the 37th move of AlphaGo. At

that moment, he says, he felt like he was playing with someone who had

a mind of his own. In this way, he opens the door to a dierent under-

standing of articial intelligence, which in categories such as beauty actu-

ally makes people not opponents, but equal (sparring) partners. e

human, in this case the AlphaGo team became the creator of the AI.e

rise in searching activity in all three elds– ML, AI and NN– suggests

that this communication has struck a chord. Google relies on a narrative

where humans and machines are not adversaries, but partners in a com-

petition. ey are successful when humans become creators.

3 From the Question to the Data

Google shifted the emphasis somewhat with the next expansion stage

of the Go game: In October 2017, Google released AlphaGo Zero, a new,

much more powerful version of DeepMind’s Go software. What was new

about it was that it didn’t learn from existing games, only from playing

against itself. Within three days, AlphaGo Zero had already surpassed its

predecessor’s playing strength in defeating Lee Sedol. By training an AI

solely by itself and foregoing human expertise, Google broke new ground

(Silver and Hassabis 2017). e release was not widely publicized in the

media, but it caused a stir beyond the professional community. In search

queries, the echo of this resonance can be traced in the form of short

search spikes for all three terms– AI, NN and ML.e narrative here is

still the same as in the rst competition against Lee Sedol, the interplay

of man and machine, but in this step the created has already conquered a

bit of autonomy from its creator.

Even if the Gartner Hype Cycle suggests otherwise: Ngram’s evalua-

tion shows that trends can recur. e constellation around the year 2010

makes this clear. AI sinks to an interim low in 2008, before a trend rever-

sal sets in and the number of mentions begins to rise again. e trend is

similar, only slightly delayed, for NN, which reaches its interim low in

2012 before the number of mentions also starts to rise again. e begin-

ning of the new decade marks a turning point: interest in AI and NN

returns, while ML begins a steep rise.

At this point, it’s worth taking a look at Google Trends. A search intent

is to be evaluated dierently than a mention in a publication. erefore,

a direct comparison of the two tools is not possible. Nevertheless, Trends

provides a good indicator of the development of interest in the topic area

(see Fig.3.6). My hypothesis is that Ngram is more likely to represent

engagement in the professional community, while the trend in Google

Trends suggests broader interest among the general public. Trends has the

e picture can be completed by a search on Google. Here, “Machine Learning” comes up with

147 million results, “Articial Intelligence” with around 136 million results, while “Neural

Network” is far behind with only 24 million results (search from 16 April 2020. Search settings

without further lters, i.e. “any language”, “any time” and “all results”). It is not possible to say

exactly where the peak in demand for the term “Neural Network” in January 2020 came from. It

may be related to Google’s launch of Flax. Flax is a Neural Network Library for JAX, a library for

high-performance machine learning.

H.-W. Eckert

Fig. 3.6 Machine learning is currently the most searched AI term. Interactive

graphic at www.data- storyteller.de. (Source: Google Trends)

merit of reecting real-time reactions, while Ngram documents develop-

ments with a time lag determined by the publication process.

For both Ngram and Trends, the numbers increase from 2010/2011

onwards for all terms except CY (here a slow increase is noticeable for the

specialist audience, while for the wider public the term is irrelevant). ML

has already overtaken AI in the specialist audience in 2013, but in the

broader public (trends) this is only the case two years later. Further devel-

opments also show that the professional world sets the topics, which are

then received by the public with a delay. Mentions of ML begin to

increase signicantly in Ngram as early as 2012, while in Trends this

eect only becomes visible from 2015 onwards.

3 From the Question to the Data

Looking at these longer cycles around AI as a topic, you see sev-

eral things:

• Trends can return– as evidenced by the resurgence of interest in AI,

ML and NN from 2010.

• e driver of the AI topic is the professional world. e changes in

trends are– despite the known delays– rst visible on Ngram before

they appear in Trends.

• In Ngram, a new peak of interest in the topic of AI is emerging (in the

terms AI, ML and NN), while in Trends the curves have already peaked

or passed. If the professional community is the driver of the topic, it is

likely that we are heading for a new hype.

For those who want to occupy relevant topics and shape an agenda, an

analysis of such correlations is helpful. Overall, the topic area shows how

developments can be analysed and predictions made for the further suc-

cess of topics using relatively simple means via tools such as Ngram and

Google Trends. It also shows that a story only has a truly far-reaching

resonance if it docks onto a successful narrative.

3.3.4 Context andChange ofMeanings

To nd out keywords that map a topic or at least partial aspects of it,

expert knowledge is a good start. But this is not always enough. Word

meanings are too diverse, depending on the culture, the social group and

the context in which they are used. Tools such as NeuroFlash, presented

in Sect. 3.1.2, support the analysis of word usage by evaluating millions

of sources. ey help to form clusters of related words and to discover

association elds.

When placing advertisements, selecting search terms and developing a

content strategy, such procedures are important. Google was one of the

rst commercial users with AdSense. e software made it possible to

automatically place an ad that matched the topic of a website. IBM used

such a process in Watson’s use in Jeopardy in 2011. ey also play an

important role in voice assistants from Alexa to Siri. Search engine tools

as well as content marketing tools oer such methods (Sect. 3.3.2).

H.-W. Eckert

ose who analyse developments of key words should be aware that

the meanings of these words change. e context plays a decisive role

here. is can be seen well in the example of the word “crisis”. In 2008,

there was a direct connection to the bankruptcy of the investment bank

Lehman Brothers and the nancial crisis. When the nancial crisis

reached the EU economy in 2009, there was then also talk of the euro

crisis and economic crisis. In 2015, on the other hand, the term crisis was

used particularly frequently in connection with the discussion about ref-

ugees. Five years later, in 2020, crisis is again closely associated with the

outbreak of the pandemic. Software developer and open data analyst

Johannes Filter has analysed this change in context around “crisis”. To do

this, he fed 13 million comments from ten years of a German news site

into his database. He then used machine learning to evaluate the texts.

His study shows how changeable terms are, even over a relatively short

period of time, and highlights the importance of contextual information

in the analysis (Fanta 2020).

A data team from the Süddeutsche Zeitung has taken on a much lon-

ger period of time with the Bundestag minutes: 70years, 213 million

words from over 4200 sessions. Such context analyses are only possible

on the basis of algorithms, neural networks and automated processes.

is involves translating words into numbers and relating them to neigh-

bouring words. e process is called word embedding. It is a collective

term for a number of language modeling techniques in natural language

processing in which words are mapped to vectors of real numbers. In this

case, Word2vec was used, a technique from Google that builds vector

spaces with several hundred dimensions. e trick now is to bring this

high-dimensional data back into a two-dimensional format that humans

can understand. With the help of these methods, the major lines of dis-

cussion in the Bundestag debates can be traced and interpreted. It

becomes apparent, for example, that the debate on climate change began

relatively late in the mid-nineties. Initially, the focus was on the destruc-

tion of the environment, and it was not until later in the 2000s that

concrete consequences such as species extinction and resource scarcity

came into play. e role of humans (man-made, climate crisis) has only

become a stronger focus of the debates in the last two legislative periods

since 2013 (Schories o. J.), see Fig.3.7.

3 From the Question to the Data

Fig. 3.7 Word meanings change. The Süddeutsche Zeitung has investigated this using the example of climate change

debates in the Bundestag. (Source: Schories o. J., own representation)

H.-W. Eckert

Such procedures go far beyond the procedure of mere word counts

shown in Sect. 3.3.3. Of course, the mere accumulation of a word does

not tell us much about the context in which it was used. e authors

make this clear with the example of the term “environment”: with the

entry of the Greens into the Bundestag in 1983, “environment” was used

in connection with nature conservation and biodiversity. In the fties

and sixties, on the other hand, “environment” was understood in the

sense of the world around us.

3.3.5 Setting Tomorrow’s Topics Today

We have seen how digital channels can provide insights into topics and

trends. And we have seen a number of tools that can be used to analyze

this. But how can we specically identify topics that are relevant to my

company’s customers? at is the core question of any company that

addresses its target groups via relevant content.

e Frankfurt-based start-up Pythia-ai (https://www.pythia- ai.de/ –

the reference of the name to the High Priestess of Delphi is probably not

entirely coincidental) has dedicated itself to the analysis of trends:

Whether tness trends or product range policy at a drugstore– Pythia

positions itself as a new oracle that uses articial intelligence to process

enormous amounts of data and derive predictions from it.

e company examines Google and Amazon searches as well as other

data sources for new trends. e start-up promises to create trend analy-

ses for upcoming market demands and claims to be much more precise

and faster than classic market research. Pythia’s algorithms take over the

collection, control and evaluation of data. e drugstore chain Rossmann,

for example, uses the tool to optimize its product range. So far, the tool

has only been responsible for selected products in the range, including

cannabidiol oil (CBD), which Rossmann says has become a sales driver.

In addition to Rossmann, other retailers, a shirt manufacturer, nancial

companies, agencies, and publishers are among the customers who use

Pythia to nd content and topics (Schobelt 2020).

ose who want to understand exactly what data ows in, how it is

collected, combined and interpreted, should consider in-house develop-

ment. It does involve signicantly more work to take the process into

3 From the Question to the Data

your own hands. However, this is the only way to ensure that the selec-

tion of data and its interpretation are also relevant to your own research

question. Moreover, with the exception of Pythia, there is no prediction

function for future trends and topics.

Of course, every topic should be in demand accordingly. To achieve

this, companies should deal with a few questions:

• What topics did we miss?

• How does my target group talk about a topic?

• How are the themes related?

• How relevant is my topic to the target group?

• Is relevance increasing or decreasing?

e content marketing tools in Sect. 3.3.2 are suitable for monitoring.

However, they oer little insight into the source situation and analytics.

ey do not allow you to create a topic map and, above all, they do not

have a prediction function.

In order to nd relevant trends and topics for their customers, com-

munication and marketing managers best proceed in three steps:

1. Identify trends and themes: Trends and themes can be derived from

the available content of the observed channels. However, the available

material is so large that it is possible to draw valid conclusions from it

using automated evaluations. So you need a suitable technology to

analyze the data. Usually, articial intelligence is used here to identify

and cluster topics. is can be done with or without human interven-

tion. ese processes are then called supervised or unsupervised. In

the latter case, the network nds patterns in the sources independently

and without any external input. With the use of unsupervised learn-

ing you make sure not to miss any topic due to limitations of your

own view. However, to train such networks, you need an enormous

amount of data. e AI then nds thematic relationships on its own

and can cluster them. In the other case of supervised learning, a

human looks at it and provides categories and topic areas. Such meth-

ods are more common today and less data intensive. However, they

have the disadvantage that certain topics and trends may not be dis-

covered in the rst place due to the human focus

H.-W. Eckert

Developments in social networks are an important and quick indica-

tor of future success. Relevant topics and trends emerge in social net-

works such as Twitter, Facebook, Instagram, TikTok, YouTube, and so

on. Companies can analyze these developments and use them to make

predictions for future topics.

To identify content with broad mass appeal, it makes sense to com-

bine various factors: socio-demographic data, overarching content

themes and interactions. With an intelligent combination of these

data sets, consumption patterns can be identied and linked to the

target groups of one’s own company.

2. Allocate topics and trends to the company’s focal points: At this point,

there is a matching between the topic areas that emerged from the

analysis of the sources and the company’s topic clusters. e aim here

is to determine which topics can play a role for the company. is step

requires a deep understanding of the company’s topics. A machine

would quickly reach its limits here.

3. Predict life cycles of topics: e selected topics are weighted according

to future relevance. e question here is: Is the relevance of the topic

increasing or decreasing and over what period of time will it develop.

is allows the company to prioritize which content should be pro-

duced and played out at what time. Mathematical models can be used

for the prediction, which make statements about the future relevance

on the basis of stochastic procedures using the past development.

3.4 Where Silence Is Golden:

Uplift Modelling

Every year just before Christmas I get mail from donation organizations.

Especially from those I have donated to before. ick letters, in which

the special meaning of this donation is colorfully described to me.

Sometimes there are give-aways like seeds, wooden spoons and other

gimmicks. You don’t just throw something like that away. But what does

such a mailing achieve? From the organization’s point of view, it’s clear:

attention, consternation and a reach for the wallet.

3 From the Question to the Data

Most of these organizations are masters of tactical storytelling: an indi-

vidual story, often a child in a life-threatening situation, a rescue inter-

vention through the organization’s eorts. And I can make it all happen

by making a small contribution from my wallet. Good conscience is as

simple as that.

But they don’t always understand the situation their addressee is in.

For me, this mail led to the fact that I am annoyed that my money does

not reach the children, but is used for postage, colorful yers and gim-

micks. An email as a little reminder that now would be the time to donate

something again would be perfectly sucient for me. Am I the only one

who thinks like that?

is question is addressed by the so-called uplift modelling (Table3.1).

e procedure is used when I want to know whether the use of a com-

munication measure causes a change in behaviour. is by no means only

applies to donation organisations. e procedure is used above all in the

retail trade and by providers of telecommunications and electricity. e

aim is to understand what eect my communication has and whom it is

better not to address. Electricity or telephone customers often use it to

discourage customers from switching providers. But it can just as easily

lead to a customer receiving the impulse to look into their contract pre-

cisely because of the letter and then discovering that there are much

cheaper competitors. With this customer the action would have approxi-

mately the same eect as the donation mailing with me. He is then gone.

Uplift modeling distinguishes two things:

• What happens when I approach the customer?

• What happens if I don’t approach the customer?

Table 3.1 Typical uplift modelling use cases according to Thurber

Use case Destination Procedure

Telephone customer Do not migrate Upgrade offer

Patient Get healthy Treat

Voter Choose Message

Donor Donate Appeal for donations

Candidate Accept offer Switching bonus

End customer Purchase Special offer

Source: Thurber (2017)

H.-W. Eckert

e two questions result in a constellation of four address possibilities

(Fig.3.8):

(a) “Do Not Disturb.” ese are all those for whom a promotion does

the opposite of what it was intended to achieve. So the customer who

switches providers when they get an oer to stay.

(b) “Lost”: Here it doesn’t matter whether the person receives a pulse or

not. He will not respond in any case. e mailing campaign is in vain.

impulse or not. He will react in any case. So here, too, futile actionism.

(d) “Persuadable”: the only case where the communication activity is

worthwhile.

An incremental procedure can now be used to determine the right target

group: First, a selection group is created, which is divided into a cam-

paign group and a control group. is is important to get feedback on the

dierent eects that arise without and with action.

Fig. 3.8 Uplift modelling helps to answer the question whether the use of a com-

munication measure causes a change in behaviour

3 From the Question to the Data

In the further procedure, there are several methods. ere are essen-

tially three main approaches here: the two-model approach, the class

transformation approach and direct modelling (see Gutierrez and Gerardy

2016). ey dier mainly in how they compute the probabilities of the

behavior of the campaign group and the control group. Crucial in our

context is the realization that sometimes I am more successful if I don’t

tell my story to everyone.

3.5 Towards aData Strategy

Every company has its specic culture, its market and customers, and its

current issues. Data helps answer these questions and make the right

decisions. Once the right question has been formulated, a strategy can be

developed to tap into the appropriate data. is data strategy must be

developed by each company based on its specic constellation of exper-

tise, sales channels, budgets, available data and the company’s maturity

level in terms of data culture.

Successful data strategies are an integral part of overall business strat-

egy. ey establish common and repeatable methods, practices, and pro-

cesses to control and distribute data across the enterprise. When the

entire organization is involved from the beginning, they can advance

their data-driven approach.

But it seems that many companies, especially in Germany, are not yet

ready. Only 26% of data teams in German companies see themselves in a

position to draw the required insights from their data. is is according

to the study “Data Strategy and Corporate Culture” by Exasol, which was

published in February 2020. e report is based on a survey of more than

2000 data strategy decision-makers in four key markets: the UK,

Germany, the US and China. On a global level, this compares to 32%.

However, this also highlights that two-thirds of all companies still feel

unable to use their data properly or have a precise idea of what data they

want to use and for what purposes.

e authors of the study are also surprised that many companies ask

the second question before the rst: Namely, they are prioritizing where

they store data and make it available in their business intelligence

H.-W. Eckert

systems– in the cloud or on premise. is is certainly an important step

towards an open data culture, where everyone has access to the same data

set (Exasol 2020). But the real question of what answers they are looking

for in the data does not seem to be on the minds of many data manag-

ers either.

Martin Szugat, for example, shows how they can proceed with his

company Datentreiber. With data strategy design, Datentreiber helps

everyone involved to ask the right questions, to identify important stake-

holders in the process, to dene goals in an interdisciplinary team and

then to develop strategies for collecting and evaluating the data. A free

toolset is available for this purpose. Datentreiber’s canvases can be used to

develop the various elds of a data strategy. Twelve canvases help identify

the topic areas and ways to address them. ey cover strategic questions

about growth horizons and value chains as well as the identication of

concrete use cases and the development of a customer contact point anal-

ysis, for example. A sample procedure shows how a marketing strategy

can be developed on the basis of data.

From one use case, dierent use cases and their benets are identied.

For example, if I want to optimize my customer journey as a company,

the individual customer contact points, such as social media posts, blog

entries, and emails, can be analyzed. On this basis, it is possible to deter-

mine on a case-by-case basis what data the company already has about

these, what potential the contact point oers and with how much eort

an optimization can be implemented. is provides the basis for priori-

tizing the next steps. Only at the end of these steps can the data sources

and tools for preparing the data be identied.

e procedure makes it clear how important it is to have a clear and

jointly developed focus at the beginning. Without a clear focus, the com-

plexity of the next steps increases rapidly and bears the risk of getting

bogged down. In order to avoid this and to successfully implement data

projects, a common understanding is needed about which goals are being

pursued, which data is needed for this, how it is to be interpreted, and

how it contributes to the goals (Klaus 2019 describes which prerequisites

are needed for this in marketing and how important people are in this

https://www.datentreiber.de/methode/#canvas

3 From the Question to the Data

process). A common data culture is therefore needed in the areas involved,

on which such questions can ourish. is can only succeed if all those

involved from IT and the business departments have agreed on common

perspectives and claried their respective roles. An important step towards

data culture is the democratization of data. It gives employees at all levels

access to data insights that are relevant to their roles. is enables employ-

ees to make better-informed decisions and nd new insights. is drives

a cultural shift, with every employee contributing to data analytics,

embedding the data strategy in the business.

References

Aaron J (2019) Valuable insights drive growth for online retailer. Station, 10.

Februar. https://www.station10.co.uk/case- studies/valuable- insights- drive-

growth- for- online- retailer. Accessed: 4. Mai 2020

Anton M (2019) Kundenzentrierung & Künstliche Intelligenz: Erfolgshebel in

der Neuausrichtung einer Traditionsmarke. Gpredictive, 2. Dezember.

https://blog.gpredictive.de/kundenzentrierung- k%C3%BCnstliche-

intelligenz- erfolgshebel- in- der- neuausrichtung- einer- traditionsmarke.

Accessed: 4. Mai 2020

Berger R (2015) Die digitale Zukunft des B2B-Vertriebs. München

Economist (2020) It has never been easier to launch a new brand. Economist,

23. Januar. https://www.economist.com/business/2020/01/23/it- has- never-

been- easier- to- launch- a- new- brand. Accessed: 4. Mai 2020

Exasol (2020) Exasol-Studie: Nur 32 Prozent der Daten-Teams können die

Erkenntnisse gewinnen, die ihr Unternehmen für eine bessere Entschei

dungsndung braucht Pressemeldung Exasol, 13. Februar. https://www.exa-

sol.com/de/company/newsroom/news- and- press/exasol- studie- nur- 32-

prozent- der- daten- teams/. Accessed: 4. Mai 2020

Fanta A (2020) Die Diskursmaschine. Netzpolitikorg, 20. Mai. https://netzpoli-

tik.org/2020/die- diskursmaschine/. Accessed: 19. Juli 2020

Fukushima K, Miyake S (1982) Neocognitron: a self-organizing neural network

model for a mechanism of visual pattern recognition. In: Competition and

cooperation in neural nets. Springer, Berlin, pp S267–S285

Gutierrez P, Gérardy J-Y (2016) Causal inference and uplift modeling. JMLR:

workshop and conference proceedings, 67/2016, S 1–13. https://proceed-

ings.mlr.press/v67/gutierrez17a/gutierrez17a.pdf. Accessed: 4. Mai 2020

H.-W. Eckert

Heßler M (2017) Der Erfolg der “Dummheit”. NTM 25:1–33. https://doi.

org/10.1007/s00048- 017- 0167- 6

Kantar (o. J.) Ad testing. millwardbrown.com. https://www.millwardbrown.

com/mb- global/what- we- do/advertising/ad- testing. Accessed: 4. Mai 2020

Klaus L (2019) Data-Driven Marketing und der Erfolgsfaktor Mensch. Springer,

Gabler, Wiesbaden

Kreye A (2018) Macht Euch die Maschinen untertan. Süddeutsche Zeitung

Edition, Münche

Lin Y, Michel J-B, Lieberman Aiden E, Orwant J, Brockman W, Petrov S (2012)

Syntactic annotations for the Google books Ngram corpus. In: Proceedings

of the 50th annual meeting of the association for computational linguistics,

Jeju, Republic of Korea, pp S169–S174

Lordick M (2016) Transhumanismus: Die Cyborgisierung des Menschen.

Zukunftsinstitut, September 2016. https://www.zukunftsinstitut.de/artikel/

transhumanismus- die- cyborgisierung- des- menschen/. Accessed: 4. Mai 2020

Mall JC (2020) Improve your brand positioning like Volkswagen. Neuroash,

15. Januar. https://neuro- ash.com/improve- your- brand- positioning-

volkswagen- neuro- ash- example/. Accessed: 4. Mai 2020

Manhart K (2018) Was Sie über Maschinelles Lernen wissen müssen.

Computerwoche, 19. Juni. https://www.computerwoche.de/a/was- sie- ueber-

maschinelles- lernen- wissen- muessen,3329560. Accessed: 16. Apr. 2020

McCarthy J, Minsky ML, Rochester N, Shannon CE (1955) A proposal for the

Dartmouth summer research project on articial intelligence, 31. August.

https://web.archive.org/web/20080930164306/, http://www.formal.stan-

ford.edu/jmc/history/dartmouth/dartmouth.html. Accessed: 4. Mai 2020

Peel S (2019) Walking the walk. E Week, 24. Oktober. https://www.youtube.

com/watch?time_continue=971&v=rbT8TqBUgOs&feature=emb_logo.

Accessed: 3. Sept. 2020

Rentz I (2019) Die Rückkehr des Gänsehaut-Faktors. Horizont, 19. Dezember.

https://www.horizont.net/marketing/nachrichten/brand- vs.- performance-

die- rueckkehr- des- gaensehaut- faktors- 179822. Accessed: 4. Mai 2020

Schobelt F (2020) Die KI sortiert das Regal: Wie Rossmann Künstliche

Intelligenz einsetzt. OneToOne, 14. Januar. https://www.onetoone.de/

artikel/db/915861frs.html. Accessed: 4. Mai 2020

Schories M (o. J.) So haben wir den Bundestag ausgerechnet. Süddeutsche

Zeitung. https://projekte.sueddeutsche.de/artikel/politik/so- haben- wir- den-

bundestag- ausgerechnet- e893391/. Accessed: 15. Okt. 2020

3 From the Question to the Data

Silver D, Hassabis D (2017) AlphaGo zero: starting from scratch. Deepmindcom,

18. Oktober. https://deepmind.com/blog/article/alphago- zero- starting-

scratch. Accessed: 12. Juli 2020

Trommsdor V, Becker J (2009) Verfahren des Werbemittel-Pretesting. In:

Bruhn M, Esch F-R, Langner T (eds) Handbuch Kommunikation. Gabler

Verlag, Wiesbaden, pp923–938

Wiener N (1948) Cybernetics. MIT Press, Cambridge

H.-W. Eckert

From Data toStory

Abstract Data is the content of stories. In addition, the proliferation of

digital channels is giving rise to a new form of storytelling. More than

ever, visual elements are becoming the hook and anchor of stories.

Especially in a society ooded with stimuli, strong visuals are gaining

weight. Journalism has shown how stories can be developed using data.

In the meantime, these practices have also arrived in companies. If you

want to turn data into stories, you need a team with very dierent skills.

Communication and marketing managers are therefore well advised to

form networks and develop a common data culture.

4.1 Data Journalism: WithWikileaks

totheBreakthrough

Anyone who has ever tried to lter out important information from an

Excel list can imagine the task the journalists faced when they received a

document with 92,201 lines from Wikileaks. On July 25, 2010, Wikileaks

published the war diary of the war in Afghanistan and made the complete

part of Springer Nature 2022

H.-W. Eckert, Storytelling with data, https://doi.org/10.1007/978-3-658-38555-2_4

documents available in advance to selected journalists from the Guardian,

Der Spiegel and the NewYork Times.

e evaluation of the document, which listed events of the war, hap-

pened under great time pressure, as Wikileaks was about to make the

entire documents publicly accessible. e journalists’ information advan-

tage melted away in a short time. At the same time, they were faced with

the task of ltering out militarily relevant information, especially so as

not to endanger informants and NATO troops. e majority of the doc-

uments contained frontline reports from the years 2004 to 2010, so a

thorough analysis was required, because any mistake would put people’s

lives at risk.

Editing such a huge Excel list was enormously time-consuming, if only

because each save operation took a long time. e data was also inconsis-

tently formatted, with many rows simply blank. Scrolling did not provide

much insight given the quantity. e journalists were faced with a seem-

ingly impossible task.

But a bit of luck is involved. Harold Frayman, editor at the Guardian,

had already gained experience with structuring data with his colleague

John Houston and developed an internal database that could process

such large amounts of data faster and easier than Excel. Journalists could

transfer the data to this and search for keywords and events there.

For example, the journalists ltered out reports on booby traps and

evaluated the approximately 7500 reports on explosions from 2004 to

2009. Using this data, the journalists were able to analyse the develop-

ment of attacks over time, by region and by destructive force. With the

help of a developer, they added the coordinates of the attacks and inserted

them into a map. A graphic designer helped them to prepare the data for

the newspaper (see Rogers 2010a, b).

e Wikileaks revelations may not have been the birth of data journal-

ism, but this was the rst time it was able to impressively demonstrate its

relevance. Never before had such large amounts of data with even

remotely similar political implications come to light. And never before

had it been possible to distil stories from such a large amount of data.

With this form, journalists opened up access to stories that would not

have been visible by conventional means.

H.-W. Eckert

In this case, data itself is the subject of the story being told. is char-

acterizes data journalism and also the many other data stories we tell in

the corporate context, which form the subject of this chapter (see Fig.4.1

and the explanations in Sect. 1.4).

e Wikileaks example vividly illustrates the challenges authors face

when ltering stories from large amounts of data:

• Large amounts of data cannot be handled without machine assistance.

• In order to exploit the data, it must be structured and standardised.

• e data (in our case the times of explosions of booby traps) are

enriched by further data (geo-coordinates) in order to gain further

insights or to be able to present them better.

• e analysis of the data requires know-how among the employees and

a corresponding data infrastructure in the company.

• In many cases, the main narrative thread is determined by graphic ele-

ments, in our example an interactive map. e preparation in turn

requires own skills in the eld of data visualization in order to make

Fig. 4.1 Storytelling with data: in this chapter, data itself is the subject of the

story being told. (Source: Own illustration)

4 From Data to Story

the important ndings visible and to leave out everything unimport-

ant without exposing oneself to the suspicion of manipulation.

Simon Rogers, then news editor of e Guardian, recalls the Wikileaks

period: “It really started with a simple idea: what if we could publish data

in a format that would be easier for others to use? … We launched with

200-odd datasets, stored on Google drives because I couldn’t get any

resources for a database. at had the weird side-eect of making our

work very easy for others to replicate. We were the rst blog about data

anywhere in the mainstream media. … rough a combination of big

stories– the WikiLeaks war records, the 2011 riots coverage and the MPs’

expenses crowdsourcing – it really took o. Suddenly, there was data

everywhere, and we explained it and made it more available” (Barr

etal. 2019).

e triumph of data journalism came up against classic journalism,

whose credibility was questioned at the time. Ten years ago, the represen-

tatives of the new discipline were therefore able to compete with the con-

viction that they could deliver comprehensible and objectiable results

with access to sources and tools.

e successes achieved by Wikileaks spurred the journalists on to open

up new sources. Once the technical prerequisites had been created and

the journalists had gained initial experience with analysing and process-

ing, they set about collecting further large amounts of data: from oces,

authorities, cities, mobile phone providers, weather stations, satellites,

etc. On this basis, stories about corruption, the East-West divide, climate

change, forest res, etc. could be developed.

Data journalism was fuelled by the growing possibilities to gain access

to data. e democratization of data access was advanced by the Open

Data movement. e idea emerged as early as the late 1950s and origi-

nally served to facilitate the sharing of data for scientic use. An

e Guardian has published its data stories in a dedicated collection: https://www.theguardian.

com/data. Here are a few more selected examples to give you an idea of the breadth of the genre:

e topics of the Bundestag over 70years from 1949 to 2019 have been analysed by the data team

of the Süddeutsche Zeitung, identifying focal points and shifts: Schories (2020). e evaluations

by Biermann (2011) and Valentino-DeVries etal. (2018), for example, show how movement data

and good contextualization can be used to learn a great deal about a person. e spread of Fake

News based on Macron’s alleged homosexuality is analyzed by Hamann (2017). Casselman and

Dougherty (2019) report on the practices of investors in the US real estate market.

H.-W. Eckert

important milestone was the 1995 report of the National Academy of

Sciences in the USA, which called for an international exchange of data

on changing the world and the environment (National Research Council

1995). Closely related to this is the open government approach, which

combines transparent governance with freedom of information and citi-

zen participation. e city of NewYork, for example, began early on to

make its data available to everyone. ere, the Mayor’s Oce of Data

Analytics (MODA) and the Department of Information Technology and

Telecommunications (DoITT) jointly form the Open Data team. As a

hub of analytics in the city, MODA advocates for the use of Open Data

in citywide data analytics and in the community. e website provides

guidance on how to use data and an overview of current projects (NYC

Open Data 2020).

Politicians in Germany have also launched initial initiatives to enable

access for citizens and companies. For example, the direct federal admin-

istration was obliged to make electronically collected data available as

Open Data (“Open by Default”) by default from 13 July 2018 at the

latest: freely accessible, free of charge and machine-readable. us, topics

such as health, mobility, air quality, weather, water levels and radiation

can be accessed on portals, e.g. at the Federal Statistical Oce as well as

the statistical oces of the Länder and many other authorities and admin-

istrations. But much remains to be done. Valuable data treasures such as

trac data and geodata are not yet publicly accessible.

Open data helps to make administration and government more trans-

parent and to improve citizen participation. It can also become a valuable

source for companies to develop new business models based on data.

ey can use public data to create new value chains as well as to expand

existing ones.

Data journalism beneted from these large, publicly accessible data

pools. In the early days, the discipline consisted mainly of the visual prep-

aration of large amounts of data, which testied to the pride of those who

compiled and prepared this data. It often lacked a narrative element such

Cf. on the situation in Germany Bildesheim (2019). A very good overview of publicly accessible

data can be found on Github (2020). Proponents of Open Data see data as freely available public

property. Since this material is of course also of economic interest, conict arises here with those

who wish to derive benet from the exploitation of the data.

4 From Data to Story

as a main character to walk you through the story. ese were stories that

stood out primarily because of the way they were graphically prepared.

Much has changed since then. e joy of the interactive tools soon

gave way to the disillusionment that the reader does not use them. is

treatment also faced criticism for not giving the reader help in explaining

the data. Today’s data stories are committed to linear and journalist-led

storytelling. Visual editing relies less on charts and more on interactive

graphics. e linear narrative form combined with many visualizations in

digital formats also earned the genre the name “scrollytelling.”

e appeal of the new thanks to sensational visuals has now worn o.

is has certainly led to a more mature approach to the genre. Above all,

however, the big data companies have also discovered the topic. e sec-

ond edition of the “Data Journalism Handbook”, for example, is spon-

sored by Google, among others. And Simon Rogers, who was part of the

founding team at the Guardian, is now– after a brief stopover at Twitter–

at the Google News Initiative.

Teams of data journalists are now established in many media houses–

from Bayerischer Rundfunk to DIE ZEIT, from the Economist to the

NewYork Times. Interdisciplinary work is a must here. According to the

self-image, the journalist is also the one who develops the questions and

classies the answers. But he needs helpers. Whereas the classic journalist

falls back on his sources, i.e. consults contacts and tracks down informa-

tion himself, the data journalist is dependent on a team.

Organizationally, data journalism has now become part of everyday

editorial life: Whereas in the early days it was mainly small, specialized

teams that worked on topics in isolation, today’s data journalists are part

of networked teams. When processing very large amounts of data, they

increasingly work together with other departments of the medium or– in

the case of major revelations such as the Panama Papers– with other media.

But even the most successful visualization does not necessarily tell a

story. For data to become relevant to us humans, it needs a context and a

protagonist. Analysis, visualization and story belong together. Only by

embedding them in narratives and personalizing them do data analyses

acquire an emotional component that is crucial for their reception.

e already available articles of the second version of the Data Journalism Handbook are evidence

of this.

H.-W. Eckert

Today, the focus is more on the human dimension of the story: “Now

we amplify the stories we nd in data by collaborating with specialist

reporters to put human voices at the center of our stories,” reports

Caelainn Barr, data projects editor at the Guardian (Barr etal. 2019).

Ben Casselman, editor of the NewYork Times, describes this approach as

follows: “e best stories almost always emerge from talking to people,

whether they are experts or just ordinary people aected by the issues we

write about. ey’re the ones who pose the questions that data can help

answer, or who help explain the trends that the data reveals, or who can

provide the wrinkles and nuances that the data glosses over. … At the end

of the day, data isn’t the story; people are the story” (Casselman 2019).

People remain the central actors in data stories.

e authors of the report on the real estate rental market in Germany

in the Süddeutsche Zeitung, for example, take this insight to heart. ey

link the data story with concrete protagonists– here, for example, the

Riedel family (Fig.4.2). e family exemplies the 44% of people who

spend more than 30% of their net household income on rent. is is a

larger than average proportion of their income, which can lead to them

being in a nancially critical situation. Another protagonist, Anna Meier,

is representative of a young, well-educated woman who, as an IT consul-

tant, is quite nancially able to aord an apartment, but has been living

with her mother in Munich for three years because she cannot nd one.

us, from over 57,000 responses to a survey, the abstract result becomes

vivid and concretely tangible (Beitzer etal. 2018).

4.2 Visualization: Basics, Tools

andBest Practice

e proliferation of digital channels is creating a new form of storytell-

ing. More than ever, visual elements are becoming the hook and anchor

of stories. ey are easy to produce and easy to distribute via smartphone

and computer. Especially in a society ooded by stimuli, in which every

story ghts for the limited attention of the addressee, strong images gain

weight. e construction of our sense of sight also contributes to this:

Between eye and brain runs the fastest data connection of all senses in us

4 From Data to Story

Fig. 4.2 Personal fates make the result of a survey of 57,000 data records vivid

and tangible. (Source: Sarah Unterhitzenberger/Süddeutsche Zeitung graphic in

Beitzer etal. 2018)

H.-W. Eckert

humans. Our brain has been optimally trained by evolution to recognize

visual patterns very quickly.

e combination of text and visualization facilitates the reception of

stories because it addresses dierent brain regions. Text is processed in the

left hemisphere of the brain, whereas the visual content is processed in

the right hemisphere, which is also responsible for emotional impres-

sions. e more brain regions are activated, the more intense the eect of

such a story.

e visualization of data is still a fairly young discipline. As recently as

the 1970s, Frank Anscombe, an English statistician, was still complain-

ing that his colleagues relied too much on table overviews and paid too

little attention to visualizations. At that time, graphs were still considered

rough approximations, while number series, on the other hand, were

considered accurate. He created four data series whose most important

statistical ratios (mean, variance, correlation, linear regression) were iden-

tical but diered fundamentally as soon as they were visualized (Anscombe

1973; Fig.4.3).

ese four graphs then became famous as the Anscombe Quartet

(Fig.4.4). In this way, he was able to show that visualizations were an impor-

tant step in the process of analysis, which could be used both for exploration

and for explaining the story. is is where the ability to recognize patterns

helps us. Conversely, this means: Once data is visualized in the appropriate

form, we process and understand it more quickly and intuitively.

Our fast thinking processes these signals and has recognized the pat-

terns before our slow thinking can intervene. In the psychology of per-

ception, we speak of so-called preattentive perception, in which our brain

lters and processes these sensory impressions before we become aware of

them. By means of preattentive features (see Fig.4.5), we can emphasize

similarities or highlight dierences. Shapes and colours can be used to

highlight certain things.

In addition to preattentive traits, Gestalt theory has contributed

important insights into perception. As early as the 1920s, psychologists

Max Wertheimer, Kurt Koka, and Wolfgang Köhler sought to under-

stand how people recognize patterns. is resulted in a set of Gestalt

principles that help us understand how visualizations of data are per-

ceived (Seel 2012, p.79).

4 From Data to Story

Fig. 4.3 Four data series whose main statistical ratios are identical (mean, vari-

ance, correlation, linear regression). (Source: Anscombe 1973)

Data storytelling can take advantage of this by optimizing design for

human perception. When data is visualized, the quantitative information

is encoded in shapes, color, position, and so on. Viewers must then

decode this information. William S.Cleveland and Robert McGill, in

their 1984 foundational work Graphical Perception, identied the

H.-W. Eckert

Fig. 4.4 Anscombe illustrates the role of visualizations in the analysis process

with the four scatter plots based on data series whose statistical ratios are identi-

cal. (Source: Anscombe 1973)

essential processes of human pattern recognition and described the

decoding of information contained in graphics. e study deals with a

small but important part of the whole process of graphical perception:

• of detection,

• the merging and grouping of the elements (assembly) and

• estimation and comparison (estimation)

rough numerous testing procedures, the authors have determined

which forms of graphics are best suited to decode this information and

have come up with the following elements:

4 From Data to Story

Fig. 4.5 Our brain ﬁlters and processes sensory impressions before we become

aware of them. The preattentive features help us to recognize patterns quickly.

(Source: Funke and Frensch 2006, p.131)

• Position along a common scale, e.g. scatter plot

• Position on identical but unaligned scales, e.g. multiple scatter plots

• Length e.g. bar chart

• Angle & slope, e.g. pie chart

• Area, e.g. blisters

• Volume, density and color saturation, e.g. heat map

• Hue e.g. Newsmap

Even though the range of visualizations has expanded since then, viewing

habits have changed, and interactive preparations in particular have

added another dimension, the study provides a solid foundation for tell-

ing stories with data (cf. Cleveland and McGill 1984).

Reduction to the essentials is a basic principle of presentation: Edward

Tufte, a US information scientist and graphic designer, coined the term

“data-ink ratio” for this. All useless components or those that distract

from the core message should disappear. Ink should only be used to con-

vey and interpret really signicant data.

is is in line with the journal-

istic demand for clarity, simplicity and unambiguity.

Tufte (1983). Tufte also became known for his criticism of PowerPoint: “PowerPoint is evil”. e

presentation tool determines the style of thinking and thus leads to a loss of information, see

Tufte (2003).

H.-W. Eckert

But which visualization ts the message? ere are a number of com-

mon types of graphics, which can also be combined with each other

depending on the question:

Comparison: Similarities and dierences can be shown with bar or col-

umn charts. Scaled symbols also work for understanding orders of

magnitude.

Trend: Developments can be displayed using line, column and

area diagrams.

Composition: In addition to pie and donut charts, grouped bars or even

pictograms can be used for this purpose.

Relationships: Dot and bubble plots help identify outliers and clusters,

a tree map shows dependencies, and a owchart shows move-

ment patterns.

4 From Data to Story

Distributions: Range charts, word clouds, and bubble charts convey

orders of magnitude.

Spatial Distributions: Maps, Heatmaps, and Voroni Diagrams

provide

information about spatial distributions and clusters.

Today, a variety of tools (Fig.4.6) make it possible to create infographics.

Visualizations in media such as the NewYork Times, the NZZ, Der Spiegel

and the SZ are based on “R”, an open-source programming language for

statistical data analysis and visualization, the rst version of which was pub-

lished in 2000. Today, R is one of the most important programming lan-

guages for solving statistical tasks. e availability of big data has helped the

language gain popularity alongside Python and Scala (Neumann 2018).

But even without programming skills, graphics can now be created.

e repertoire of tools includes infographics, scales, geographical maps,

concept clouds, heat maps and fever curves. Interactive functions, down-

load options for data and graphics are part of the programs’ range of

services, as are a variety of templates on which designs can be created.

Most applications are cloud-based and can be used without further

local installations. e results can usually be prepared for online and

Voroni diagrams are based on a simple but powerful concept: given a set of locations in a space,

divide that space into cells– one cell for each location. Each cell contains all the points that are

closer to that location than to any other. is form of visualization is useful to many dierent elds,

such as spatial/network analysis, pattern recognition, label placement on maps, and graphs. Voroni

diagrams are now often visualized using a javascript library cf. Rivière (2017).

H.-W. Eckert

Fig. 4.6 Graphics can be created with the appropriate tools even without programming knowledge. Interactive functions

belong to the performance spectrum of the programs as well as a variety of templates on which designs can be created,

Interactive graphics at www.data- storyteller.de. (Source: Own representation)

4 From Data to Story

social media channels. For the latter, many of them have their own templates

with the optimal formats. ere are complete free solutions such as

Chartbuilder, Dash, Google Data Studio and Qgis as well as paid versions

such as Tableau. In between, numerous tools position themselves with

free versions and paid updates. In this way, you can get a good insight

into the range of functions, the interfaces, the available templates and the

user-friendliness and then decide whether a subscription solution with

usually monthly payments comes into question.

To get an overview, it is also a good idea to take a look at the collection

of Visualising Advocacy. ere, numerous concrete use cases and suitable

tools are listed, including:

• Combine and integrate dierent data sources

• Purge a record

• Build table with text and numbers

• Comparing numbers and counting words

• Making changes visible over time

• Show things on a map

• Visualise network structures (Visualising Advocacy 2020)

rough its proliferation of apps on the Android operating system and

its web applications, Google has a large and growing inuence on data

visualization standards. Its Material Design subsidiary has set guide-

lines for data visualization based on reduction and minimalism. e

design is based on map-like surfaces and the at design approach, which

is known for its minimalism. Animations and shadows are used to rep-

resent objects as physical objects with appropriate behavior, allowing

the user to immediately see which areas contain important information

or are interactive and what that interaction will do. e company is

continuously updating its web services based on Material Design and

also provides interfaces for other developers to implement the design

guidelines. In the currently available beta version, the company gives

three guiding ideas for the design:

• Accurate: Data should be presented accurately, clearly and with integrity

so that information is not distorted.

H.-W. Eckert

• Scalable: Visualizations should be adapted to device sizes. User require-

ments in terms of data depth, complexity and nature should also be

taken into account during implementation.

• Helpful: Users should be guided in navigation and encouraged to

make their own comparisons and explorations (Material Design 2020).

e rst two points would also be signed by data journalists and UX

designers. e last design principle makes it clear that Material Design,

and thus Google, have a slightly dierent perspective and allow the user

more freedom in their own exploration here than a journalist would.

When implementing infographics, diagrams and maps, the rst thing

to avoid is mistakes in craftsmanship: e use of certain colors and color

combinations leads to the fact that people with red-green weakness or

another limitation of color perception have diculty reading the graphic.

Datawrapper has a useful function here, with which you can simulate the

common visual impairments on your own graphics right away and thus

get an impression of how the implementation aects people with

disabilities.

With black-and-white illustrations, on the other hand, representations

with multiple elements quickly reach the limits of what the eye can still

distinguish. Dierent hatchings can help here. However, if the lines are

too close together, they may create a ickering eect on the viewer, the

so-called moiré eect.

But even if you can make everything nice and colorful thanks to a

freely selectable color palette, it’s not always advisable to use the entire

spectrum. Lexie Kane, UX designer at the Nielsen Norman Group, rec-

ommends reducing the color spectrum: Rather than a colorful hodge-

podge, she advises using a few accent colors that emphasize the core

message (Kane 2018).

Not only the colors, but also the other elements of graphical represen-

tations lend themselves to standardization. is also includes key gure

formats, units, fonts, headings and visualization types. Such recurring

elements make it much easier to absorb information. e best way to

document the essential building blocks is in a style guide. If such visual-

izations play a major role in corporate communication, such documenta-

tion can also be designed in great detail. e International Business

4 From Data to Story

Communication Standards, which are provided by the non-prot IBCS

Association at www.ibcs.com, oer very helpful principles and examples.

Less is more: this also applies to interactive representations. Interaction

is used so often in digital media that people are already looking for it.

Online visualization oers the possibility for users to ask questions of a

tool themselves and thus explore the data pool. For example, the Guardian

has developed a pandemic simulator that can be used to identify various

factors inuencing the spread of epidemics (Evershed and Ball 2020). In

this case, this certainly helps to familiarise oneself with the impact of dif-

ferent inuences and how they interact. However, not every interaction

serves the cause either, but sometimes rather distracts from the actual

message. erefore, you should think carefully about what is necessary

and what is better simply left out.

After all, a visualization should always speak for itself. Because it is not

always possible to prevent it from reappearing in another context. Because

graphics can be easily copied, forwarded in messenger services and posted

on social media channels: Always remember that the representation also

works without context and that sources are deposited. en the creator

will also benet if the graphic goes viral.

4.3 Finding theHero intheData

Human references are an essential building block of a data story. How does

an abstract number become tangible in a personal destiny? In Sect. 2.2 I

briey introduced classical storytelling– the hero’s journey with the setting

out, the development of a conict, its dramatic climax and its resolution.

Aristotle described the original form of this pattern in his Poetics in

335B.C. and thus founded Western literary theory (Höe 2009).

Data provide the raw material of the story. But it is only in the interplay

of linguistic and visual interpretation that they achieve their eect. And

in more ways than one:

• Memory: Chip Heath, a Stanford professor, concluded in a study that

63% of an experimental group could remember stories, but only 5%

could recall a single statistic. It was not the graph but the story that got

the group’s attention (Heath and Heath 2008, p.242).

H.-W. Eckert

• Persuasion: In another study, researchers tested two versions of a bro-

chure for the charity Save the Children: one was based on the story of

Rokia, a seven-year-old child from Mali. e other used infographics

to highlight the plight of children in Africa. e version with the story

generated twice as much donations per capita as the one with the

graphic presentation (Heath and Heath 2008, p.166).

• Engagement: mathematician John Allen Paulos observed that when

listening to stories we tend to suspend our critical minds in order to be

entertained, while conversely when processing statistics we are more

inclined to nd a hook in order not to be deceived (Dykes 2016).

Aristotelian tragedy develops in a kind of triangular movement from the

initial situation through the development of a complication, which then

culminates in the central conict and is subsequently resolved. For

Aristotle, the whole consists of a beginning, middle and end– the classic

three-step of tragedy. e core element in his model is the so-called myth,

which can best be described as the plot or sequence of a story.

An extended form of narrative structure is provided by the German

writer Gustav Freytag in his “Technique of Drama” of 1863. His insights

are based on the analysis of ancient dramas and contemporary tragedies

by Shakespeare. Freytag extended the Aristotelian triad to a development

in ve steps. e central point here is also the climax and turning point

of history. Freytag adds to the Aristotelian model one step before and one

step after the climax:

e introduction: Here the characters are introduced and the constel-

lation is explained. In addition to the hero, his antagonist also appears

here. e introduction is followed by an initial impulse that triggers the

course of the story. is intensies as it progresses, culminating in the

climax. is is followed by a phase of deceleration, in the course of which

the resolution is prepared as the next turning point.

e resolution ends with either the death of the hero and/or the reso-

lution of the conicts and the purication of the hero.

e much more comprehensive model developed by Joseph Campbell

in his hero analysis “e Hero with a ousand Faces” in 1948 is more

suitable for epic areas in which large stories with many facets, develop-

mental steps and parallel strands can be depicted. e aforementioned

4 From Data to Story

Star Wars episodes and other Hollywood material are better suited for

this than stories about companies and products.

ese narrative structures have long since found their way into corporate

communication. Today, this technique is used in many disciplines and

long since not only for lms, but also for corporate formats (cf. Dykes

2016, p.171). e structure of speeches often follows this pattern. Many

presentations, workshops, webinars, and even project reports and board

papers today, for example, have a three-stage structure of problem,

solution, and implementation. Stories now play an important role for

strategy and change processes, because there, too, it has been recognized

how important shared narratives are for coordinated action.

Such a structure can also be applied to data stories. Here, the basic

Aristotelian pattern of the three-stage structure of a drama is the most

universal principle, suitable for a wide variety of stories. He required that

every drama must have a unied, closed plot with a beginning, middle,

and end (unity of action), be set in a single location (unity of place), and

not exceed a reasonable duration (unity of time).

Aristotle’s drama form can be translated into a three-stage structure of

a story:

1. Start: Problem/Conict

2. Middle: Insight/Climax

3. End: Solution and decision

Exploring, trying out, developing and discarding again are components of

the process that leads to the story. But the process of discovery and the story

you later make out of it are two dierent things. e two should remain

strictly separate. In exceptions, a few elements of the cognition process can be

incorporated into the story later. But only if it plays a role in the internal logic

of the story you want to tell and is relevant to the target audience. Otherwise,

it has no place in the story. In Sect. 4.4, I will give two examples to show the

dierence between the explorative and the explanatory approach.

But who is the acting person? With data it is obvious to let numbers

speak and to argue with percentages, averages and normal distributions.

But our brain looks for the human dimension and thus remembers the

story much better. e good thing is that a lot of data can be traced back

to human behavior. After all, a car’s sensor data doesn’t just tell us

H.-W. Eckert

something about its current fuel consumption or engine temperature,

but also something about the person sitting behind the wheel and press-

ing the accelerator. Any investigation, no matter how data-heavy, can be

made more concrete with insights into people’s situations. is is what

the authors of the article on the rental market just cited in Sect. 4.1 have

done: With their protagonists, they highlight typical examples of a gen-

eral development and make the concrete eects clear.

In English-language literature, which is much more strongly inuenced

by narrative approaches, such stories are often developed on the basis of

storyboards. e Walt Disney Studios rst introduced the concept in the

early 1930s for the development of animated lms. Today, the storyboard

approach provides a practical procedural model for developing stories of

all kinds, including data stories.

For the practical implementation, it is good to work with sticky notes

that are attached to a magnetic board or a ipchart with the model of the

story. In this way, dierent steps along the story can be tried out and visual-

ised. e sticky notes force brevity and conciseness. And they can be rear-

ranged at any time so that dierent ways of telling the story can be tried out

(cf. Dykes 2016, pp.170–180; Nussbaumer Knaic 2020, pp.20–21).

e pivotal point is the insight/climax of the story. Why is this insight

so relevant for the addressees, what are the implications of this insight? It

is worth investing time in drawing out this central moment and sharpen-

ing the argument well. e storyboard format helps with this. It can be

used to try out and discard dierent points. It should not take more than two

sentences to formulate this insight. If it is longer, it should be sharpened.

Let’s take the example of the shoe retailer “Seven Feet Apart” from Sect.

3.2 again. e situation– ctitious here, of course– is as follows: Two

years after the launch of the online shop, growth is attening out and sales

are shrinking despite the same marketing activities. e managing director

wants to know from the marketing manager why the current marketing

eorts are no longer really working and what measures he proposes for

further sales growth. e marketing manager analyses the data of the web-

site usage, the shop system and the access to the newsletters. After dividing

the customers into age groups and analyzing the sales of these groups, he

comes to the conclusion: e core target group is older than we originally

assumed. e key point of his presentation might look something like this:

4 From Data to Story

Step 1: From the Insight to the Initial Situation From the insight, he

develops the story backwards for the presentation to management. is

way he easily nds the right point. He sharpens the problem and arrives

at the following statement: We make too little turnover per customer.

Once the problem is set, the empathetic part follows: whether the

story hits depends largely on how it connects to the audience’s expecta-

tions and experiences. With too little context, the marketer runs the risk

of losing his audience because they don’t understand the problem

H.-W. Eckert

statement. Too much potentially dilutes the ow of the story and bores

his audience. Most importantly, he should know the desires, objections,

and resistance these people have so he can address them in his argument.

In this case, he doesn’t need deep research. He knows the management

since they started the company together and knows where the crucial

points are for the CEO (wants to become market leader in premium

online retail for shoes), his CFO (wants stable earnings) and the purchas-

ing manager (is looking for suppliers with the best margins).

e same presentation in front of a circle of marketing colleagues would

need an introduction to the specics of the fashion industry, online retail

and perhaps also the shoe market in England. It is worth putting a lot of

eort into this and nding out as much as possible about the context and

the protagonists. Knowing the desires, attitudes and motives of these peo-

ple as well as possible provides the basis for a good story.

Step 2: From the Problem to the Insight How many arguments the

marketer needs between the two points depends on the story and the

audience. Anything that doesn’t t this line, he left out. A few questions

help him choose the relevant points:

• What insights help to come to understanding or provide rele-

vant context?

• What questions from the audience can be anticipated?

• Which ndings were surprising and unexpected?

• What insights can be left out without the story suering as a

result thereof?

e structure of a storyboard quickly provides the clues as to which

arguments t in well, where some are missing, and which do not t the

story. e marketing manager decides on the following arguments:

• New customer growth is slowing (is worries the CEO).

• e repeat buyer rate is lower than expected (is surprises the buyer).

• Revenue per purchase is below the industry average (the CFO would

have asked him for this benchmark anyway).

4 From Data to Story

Step 3: From Insight to Solution After presenting the main point, the

marketing manager presents his solution approach. e basis is the real-

ization that a too young customer group was addressed. is realization

was possible after he had formed age group clusters and evaluated these

groups according to their buying behavior. is then led to the solution:

the focus in future will not be on the target group of 35–45year olds, but

on 45–60 year olds. is group has more purchasing power and also

oers growth potential because market penetration is still low here.

However, this requires not only a reorientation of communication (other

channels, new themes), but also a change in the product range (wearable

shoes, sustainable production). For this, he needs the CFO and the pur-

chasing manager on his side. Because in the latter’s experience, margins

are low here, which reduces prots. e core argument at this point is:

this customer group is less price-sensitive. So larger margins are possible.

After he has outlined these approaches and the associated growth poten-

tial (important for the managing director), he still provides a proposal for

the concrete next steps with which the implementation can begin: Develop

a new customer approach with new content on the website and a campaign

specically for the age group. Introduction of customer value as a central

control variable, development of a new product range policy, etc.

4.4 Saving Lives withData: FromCholera

toCorona

In August 1854, the people of London experienced a devastating cholera

outbreak. is was part of a cholera pandemic that was rampant through-

out the world from 1846 to 1860. e largest city in Europe at the time,

London saw a large inux of people and was the center of the industrial-

ized world with a population of more than two and a half million. e

buildings and infrastructure were not up to the task. Where a sewage

system existed at all, it transported human and animal excreta as well as

industrial waste water directly into the ames. e then slum area of

Soho was particularly badly aected. Within a week, 10% of the district’s

H.-W. Eckert

residents became infected. Stables, slaughterhouses and grease houses

lined the streets, leaving behind animal excrement, rotting uids and

other lth that could not even be discharged into the ames

through sewers.

e physician John Snow had been studying cholera for some time.

Six years before the outbreak, he had published his views on the ways of

the disease. Until then, the miasma theory had dominated medical doc-

trine. According to this, cholera spread through the air, which was not

entirely far-fetched given the stench in the city. Snow, however, was con-

vinced that the spread was not through the air, but through germs. But at

this point he could not trace the path of the disease. e cholera outbreak

in Soho gave him the opportunity to test his theory and study the ways

in which it was spread. To do this, he plotted all the infections on a map.

Every bar a dead person. is map put him on the trail. rough his

work, Snow discovered a particular cluster of deaths around a pump in

Broad Street (now Broadwick Street, see Fig. 4.7). Only workers at a

nearby brewery did not fall ill.

It turned out that the pump’s water was contaminated by sewage from

a septic tank. e map provided a detailed statistical analysis of the deaths

and led him to realize that cholera is transmitted by germs in the water.

It also allowed him to explain why the workers at the brewery didn’t get

sick: ey had their own well. Because Snow prevailed with this view and

the health authorities and the population drew the right conclusions

from it, this was the last cholera outbreak London had seen. Four years

later, in 1958, the year of John Snow’s death, one of the most signicant

hygiene measures was put in place with the construction of an ecient

new sewerage system. It still forms the backbone of London’s wastewater

management today.

Snow’s map is signicant because it enabled the breakthrough of a

medical discovery. Without visualization, Snow would not have been

able to track the germs in the water, which enabled him to disprove the

miasma theory and establish a new interpretation of the infection. It was

For detailed documentation and the map, see the John Snow Archive and Research Companion

website at Vinten-Johansen (2020). For more on Snow, see Rogers (2013) and Menden (2020).

e concept of this visualization is a so-called Voroni diagram, see Sect. 4.2.

4 From Data to Story

Fig. 4.7 Deaths around a pump in Broad Street during the 1854 cholera outbreak

in London. (Source: Wikimedia Commons https://upload.wikimedia.org/wikipe-

dia/commons/c/c7/Snow- cholera- map.jpg)

only by recording the exact locations of the dead that he was able to nd

the source of the epidemic, thus pointing the way to its eective control.

e Cholera Map has since been hailed as a major pioneer of data visual-

ization. It highlights how visualizing helps people see patterns and draw

H.-W. Eckert

the right conclusions from them. But it is much more than that, it is the

document of a scientic breakthrough.

But this visualization does not tell a story. It made it possible to derive

a story from it. It is the explorative preliminary stage of a story.

Another example shows how visualizations tell stories and bring them

precisely to the point. In view of the spread of Covid-19, Siouxsie Wiles,

a microbiologist from New Zealand, developed a graphic together with

illustrator Toby Morris and published it on Twitter on 8 March 2020. It

went viral under the hashtag #attenthecurve.

e graph shows two possible progression scenarios for the new

Corona pandemic: one with a steep rise and fall and a short time course,

and one with a atter but longer course. A dashed line marks the capacity

of the health care system. e graphic idea comes from a 2007 publica-

tion by the Centers for Disease Control and Prevention (CDC). at’s

the U.S. government’s disease control agency. e pre-Corona version

already showed the two progression scenarios with and without interven-

tion (https://stacks.cdc.gov/view/cdc/11425).

An essential element has been added in the Wiles and Morris version

(Fig.4.8): a dashed line. is marks the capacity limit of the health care

system. is gives the story a crucial twist: the steep curve clearly exceeds

the capacity limit, while the at curve remains below it. e underlying

message: As soon as the healthcare system is no longer able to treat all

those who fall ill, the death rate will rise massively. Not only that, but

when the system becomes overwhelmed, things will get worse for all of

us. e creators of the graphic then backed up the two attitudes with two

gures: An indierent man, who considers the risk to be low, represents

the rst curve. A woman who urges caution stands for the second curve.

Spurred on by the response to this visualization, Wiles and Morris

developed another graphic (Fig.4.9). It gives concrete action instructions

on how everyone can act in the epidemic and went viral on 21 March

2020 under the hashtag #stopthespread. Our brains cannot conceive of

exponential progressions, which is especially dangerous in times of epi-

demics. is is exactly where the animation comes in and makes this clear

in a very descriptive way: the graphic with dots and lines shows in the

rst run how an infection spreads exponentially. A starting point on the

left side connects via lines with three more, which in turn connect several

4 From Data to Story

Fig. 4.8 The dashed line marks the capacity limit of the healthcare system. With

this crucial twist, microbiologist Siouxsie Wiles illustrates the reason for keeping

the infection curve ﬂat. (Source: Twitter)

more points, until a whole bundle of bundles is visible on the right side.

In the second pass, the graphic shows how this chain can be interrupted

by concrete actions: “Worked from home”, “Didn’t go to that BBQ”,

“Didn’t y”, “Stayed home”. ese interventions now colour the origi-

nally pink dots and lines grey as they progress. e message: everyone can

stop the spread of the virus with their actions.

ese graphics in times of cholera and Corona could save lives, albeit

in dierent ways. Wiles’ Corona visuals on Twitter stand on their own.

H.-W. Eckert

Fig. 4.9 Concrete examples show how anyone can prevent the spread of the

virus and thus curb its exponential growth. (Source: Twitter)

ey are the story and provide the specic instructions for action. Wiles

is using her animated graphics to help change patterns of public behavior

to slow the spread of the pandemic. e graphic execution gets to the

heart of the idea behind it. e visual argument in this case is stronger

and faster than any text.

In the case of the cholera epidemic, John Snow’s map of London with

the dead provided a clue to the accuracy of his thesis of germs as the dis-

ease spreaders of cholera. In my model, the Snow graph is on the explor-

atory side– it brought John Snow to an understanding of the routes of

infection and provided him with the crucial arguments to convince local

government of his theory of infection. It is only in the context of the

medical debate about miasms and germs that it comes into its own. John

Snow needed this map as a protagonist to establish the context and con-

vince his counterparts. Only with it was he able to persuade those respon-

sible in London to build a sewage system (see Fig. 4.10 and the

explanations in Sect. 1.4).

4 From Data to Story

Fig. 4.10 The difference between exploring and explaining– from the example

of John Snow and Siouxsie Wiles’ graphics. (Source: Own representation)

4.5 Data Storytelling Is Teamwork

Journalistic practices have long since found their way into companies.

Long before the triumph of data journalism, companies began to act as

media houses. ey were fascinated by the idea of getting in touch with

target groups directly and no longer having to rely on journalists for

mediation. is approach was called corporate publishing. e term

emerged early in the twentieth century, but it gained momentum in

Germany in the 1990s. Publishing houses had the opportunity to market

their journalistic expertise to companies. Customer magazines, employee

H.-W. Eckert

magazines and editorially prepared annual reports were the agship prod-

ucts of corporate publishing (Fux 2019).

With the advent of digitalization, content marketing took over from

corporate publishing. It started with the claim to oer journalistically

prepared content on all channels accessible to companies. In addition to

the already mentioned regularly recurring publications, other, primarily

digital channels were added. And the term also made it clear that it was

now marketing, rather than the press, that was claiming leadership, and

that in addition to magazines and journals, corporate blogs, newsletters

and social media sites were now also being supplied with content.

e Wikileaks example has made it clear: if you want to turn data into

stories, you need a team with very dierent skills. Wherever data are the

sources of wisdom, a Pythia is needed to open up access to these sources and

a priest to interpret the Pythia’s words. In today’s corporate world, multiple

functions vie for this interpretive authority. e CEO or managing director

does not always give the communications and marketing managers the most

ear. Depending on the industry, market focus and corporate culture, the

power of interpretation tends to lie with IT, nance, product development,

service or sales. In B2B companies, communications and marketing leaders

rely on good interplay with sales and product development, while their

counterparts in B2C companies with strong brand orientation, broad target

groups, and large budgets can more easily take the lead. However, digitiza-

tion gives all communications and marketing leaders an important asset–

access to the customer. e availability of knowledge about customers, on

the one hand, and customers’ changing expectations of companies, on the

other, are leading to a growing conviction in many companies that innova-

tions must be driven primarily from the customer’s perspective.

In order to be successful, the communication or marketing area is depen-

dent on cooperation with other specialist areas, i.e. above all IT, product

development, service and sales. In order to speak at eye level, today it needs

a deep understanding of the data sources and their potential. is requires

dierent skills in the team than ten years ago. Of course, there is now some-

one there who can condently play the keyboard of social media channels.

But where does the content come from, who controls the campaigns, ana-

lyzes them, and networks them with other activities?

At branded companies and retailers, a performance marketing man-

ager will develop the sales-oriented campaigns. But as the example of

4 From Data to Story

100

Adidas has shown, there needs to be a counterweight on the side of brand

management that can explain which communication activities pay o for

the brand and in what way, and which can damage the brand.

e technology lead will probably be based in the IT area of the com-

pany. A good connection there is of utmost relevance. Ideally, the tech-

nology lead will be in charge of the business intelligence application, into

which all data from sales, marketing, purchasing, service, etc. ows.

Valuable insights for the planning of future campaign and sales activities

can be derived from this.

However, many companies have not yet established central data stor-

age. en it is even more important to bring together the information

from the various systems such as the online shop, CRM, ERP and the

channels managed in the communication units such as the website, social

media channels and the email system. A data strategist helps identify the

right sources and discover correlations. When tapping into source mate-

rial, it helps if someone knows scraping, which is the automated extrac-

tion of data from websites, applications or documents, and can turn it

into a set of structured raw data. For example, information can be

extracted from websites of potential customers (see Gervalla 2020). e

analyst, in turn, examines the data and identies trends, develops predic-

tions, and thus transforms data into information. To do this, he or she

must master databases and how to query them, be procient in business

intelligence tools, and also know something about visualizing data.

e data strategist or data scientist (the terms are not always used uni-

formly and also change) in turn translates the questions of his team into a

strategy for the development and preparation of the data sources, he knows

or has mastered the methods of machine learning, including, for example,

software such as Matlab or programming languages like Python and R.

Exciting new roles are emerging at the boundaries of existing disci-

plines, for example between business and technology or between data

analysis and its visualization. On the visualization side, there is the so-

called UX designer, who takes care of the most positive user experience

(UX) possible. For the UX designer, it’s all about usability and the inter-

action design of a product or service. An infographic designer is familiar

with the techniques it takes to make connections visible to the viewer,

highlighting the important things and leaving out the unimportant ones.

He or she knows the importance of scales, colors, chart types, and is also

H.-W. Eckert

101

aware of the possible manipulations that can result from truncated scales

or certain color schemes. He or she will have learned more than the UX

designer about data preparation in his or her training and will have ana-

lytical thinking, but he or she also lacks the data analysis side.

A border crosser between the two worlds of data and visualization

would be a data UX designer, for example. He or she must understand

something about data analysis and data design workow and be able to

master the common visualization tools with graphic types, color palettes,

etc. as well as data storytelling with a view to the expectations and wishes

of the customer (cf. Münster 2019).

Given the large number of possible issues, it makes sense to rst

approach the topic as pragmatically as possible. A generalist approach

helps here. Overview is initially more important than focus. It is impor-

tant to know the relevant players in the company and to develop a role

allocation with them. is means that a data strategist or an analyst does

not necessarily have to be on board in the communications team if this

role already exists in the company. And if it doesn’t already exist, you

don’t have to hire someone with that title to begin with. All you need to

do initially is develop someone in this direction and continually build the

skills within the team. Depending on the size of the department and the

distribution of tasks in the company, these functions can be distributed

among several people or bundled in one person.

Many of these developments and the resulting new requirements are

technologically driven. e skills required for this are not necessarily

among the core competencies of communications and marketing manag-

ers. ey would therefore be well advised to expand their network within

the company and create a common understanding of how the handling

of data aects collaboration with IT, sales, research and development,

and service, for example. In order to develop a data culture, a good start

would be to overcome silos and divisional thinking and work on a com-

mon understanding of the challenges. After all, the other areas are also

changing as a result of digitalization. So why not learn from each other

instead of hiding behind outdated tasks and roles?

If they use this correctly, communicators and marketers can make an

important contribution to the company’s success and thus improve their

own relevance within the interpretive structure of a company. Depending

on the company’s orientation and the role it assigns to the brand, this role

4 From Data to Story

102

will lie with the head of communications and/or brand management. is

person develops the story of the company or brand and translates it into

the language of the target groups. As a storyteller, he looks at the interac-

tion with the eye of a dramaturge: Are the themes set correctly, are the

characters right, is the story clearly drawn and does it develop? In order to

be able to do this, however, he ideally has all communication topics in

view and does not sort them into pigeonholes such as press and market-

ing, but recognizes the interaction of the disciplines as an opportunity.

References

Anscombe FJ (1973) Graphs in statistical analysis. Am. Statistician 27(1):17–21

Barr C, Chalabi M, Evershed M (2019) A decade of the Datablog: ‘ere’s a human

story behind every data point’. e Guardian, 23. März. https://www.theguard-

ian.com/membership/datablog/2019/mar/23/a- decade- of- the- datablog-

theres- a- human- story- behind- every- data- point. Accessed: 11. Dez. 2019

Beitzer H, Ebitsch S, Endt C, Öchsner T, Schories M, Zajonz M (2018)

Deutschlands Mietmarkt ist kaputt. Süddeutsche Zeitung, 26. Juli. https://

projekte.sueddeutsche.de/artikel/wirtschaft/miete- wohnen- in- der-

krise- e687627/. Accessed: 1. Juni 2020

Biermann K (2011) Was Vorratsdaten über uns verraten. Die Zeit, 24. Februar.

https://www.zeit.de/digital/datenschutz/2011- 02/vorratsdaten- malte- spitz.

Accessed: 11. Mai 2020

Bildesheim O (2019) So nden und nutzen Sie oene Daten. Computerwoche,

6. Dezember. https://www.computerwoche.de/a/so- nden- und- nutzen- sie-

oene- daten,3548143. Accessed: 6. Mai 2020

Casselman B (2019) In data journalism, tech matters less than the people.

NewYork times, 13. November. https://www.nytimes.com/2019/11/13/tech-

nology/personaltech/data- journalism- economics.html. Accessed: 12. Dez. 2019

Casselman B, Dougherty C (2019) Want a house like this? Prepare for a bidding

war with investors. New York times, 20. Juni. https://www.nytimes.com/

interactive/2019/06/20/business/economy/starter- homes- investors.html.

Accessed: 12. Dez. 2019

Cleveland WS, McGill R (1984) Graphical perception: theory, experimenta-

tion, and application to the development of graphical methods. J Am Stat

Assoc 79(387):531–554

H.-W. Eckert

103

Dykes B (2016) Data storytelling: the essential data science skill everyone needs.

Forbes, 31. März. https://www.forbes.com/sites/brentdykes/2016/03/31/

data- storytelling- the- essential- data- science- skill- everyone- needs. Accessed:

12. Dez. 2019

Evershed N, Ball A (2020) How coronavirus spreads through a population and

how we can beat it. e Guardian, 22. April. https://www.theguardian.com/

world/datablog/ng- interactive/2020/apr/22/see- how- coronavirus- can-

spread- through- a- population- and- how- countries- flatten- the- curve.

Accessed: 11. Mai 2020

Funke J, Frensch P (2006) Handbuch der Allgemeinen Psychologie– Kognition.

Hogrefe, Göttingen

Fux K (o. J.) Corporate publishing– eine denition. Mediapunk.org. https://

www.mediapunk.org/corporate- publishing- denition. Access: 29. Nov. 2019

Gervalla A (2020) Erweitern Sie Ihr Kundenwissen mit Hilfe von B2B Web Scoring.

B2B Smart Data, 3. Dezember. https://www.b2bsmartdata.de/blog/erweitern-

sie- ihr- kundenwissen- mit- hilfe- von- b2b- web- scoring. Accessed: 10. Dez. 2020

Github (o. J.) Awesome public datasets. https://github.com/awesomedata/

awesome- public- datasets. Accessed: 6. Mai 2020

Hamann G (2017) Macron ist schwul, NOT! Die Zeit, 24. Februar. https://

www.zeit.de/politik/2017- 02/fake- news- emanuel- macron- russland-

rekonstruktion. Accessed: 11. Mai 2020

Heath C, Heath D (2008) Made to stick. Why some ideas survive and others

die, Random House, NewYork

Höe O (2009) Der wahre Aristoteles. FAZ, 27. Januar. https://www.faz.net/

aktuell/feuilleton/buecher/rezensionen/poetik- der- wahre- aristoteles- 1759

205.html. Accessed: 5. Juni 2020

Kane L (2018) Designing eective infographics. Nielsen Norman Group, 12.

August. https://www.nngroup.com/articles/designing- eective- infographics/.

Accessed: 17. Jan. 2020

Material Design (o. J.) Data visualization beta. Material Design. https://mate-

rial.io/design/communication/data- visualization.html#principles. Accessed:

11. Mai 2020

Menden A (2020) Das Ende des großen Gestanks. Süddeutsche Zeitung,

3. Mai. https://www.sueddeutsche.de/kultur/stadtplanung- das- ende- des-

grossen- gestanks- 1.4895361. Accessed: 4. Mai 2020

Münster E (2019) 9 design pitfalls on the way to a successful data product.

Towards data science, 27. Nov. https://towardsdatascience.com/9- design-

pitfalls- on- the- way- to- a- successful- data- product- 6ea5a3e6842. Accessed:

16. Febr. 2020

4 From Data to Story

104

National Research Council (1995) On the full and open exchange of scientic

data. e National Academies Press, Washington, DC. https://doi.

org/10.17226/18769. Accessed: 6. Mai 2020

Neumann A (2018) 25 Jahre: Wie R zur wichtigsten Programmiersprache für

Statistiker wurde. heisede, 3. Aug. https://www.heise.de/developer/meldung/

25- Jahre- Wie- R- zur- wichtigsten- Programmiersprache- fuer- Statistiker- wurde-

4127034.html. Accessed: 9. Mai 2020

Nussbaumer Knaic C (2020) Storytelling with data – let’s practice.

Wiley, Hoboken

NYC Open Data. https://opendata.cityofnewyork.us/overview/. Accessed:

6. Mai 2020

Rivière (2017) e state of d3 Voronoi, 3. Januar. https://visionscarto.net/the-

state- of- d3- voronoi. Accessed: 23. Febr. 2020

Rogers S (2010a) Wikileaks' Afghanistan war logs. e Guardian, 27. Juli.

https://www.theguardian.com/news/datablog/2010/jul/27/wikileaks-

afghanistan- data- datajournalism. Accessed: 27. Nov. 2019

Rogers S (2010b) Afghanistan IED attacks– 2006 to 2009. SCRIBD https://

de.scribd.com/document/34850058/Afghanistan- IED- attacks- 2006-

to- 2009. Accessed: 27. Nov. 2019

Rogers S (2013) John Snow's data journalism: the cholera map that changed the

world. e Guardian, 15. März. https://www.theguardian.com/news/dat-

ablog/2013/mar/15/john- snow- cholera- map accessed: 19. Jan. 2020

Schories M (2020) So haben wir den Bundestag ausgerechnet. Süddeutsche.de

(ohne Datumsangabe). https://projekte.sueddeutsche.de/artikel/politik/so-

haben- wir- den- bundestag- ausgerechnet- e893391/. Accessed: 11. Mai 2020

Seel NM (2012) Gestalt psychology of learning. In: Seel NM (ed) Encyclopedia

of the sciences of learning. Springer, Boston

Tufte E (1983) e visual display of quantitative information. Graphics Press,

Cheshire CT, Edward R.Tufte

Tufte E (2003) PowerPoint is evil. wired, 9. Januar. https://www.wired.

com/2003/09/ppt2/. Accessed: 7. Mai 2020

Valentino-DeVries J, Singer N, Keller MH, Krolik A (2018) Your apps know

where you were last night, and they’re not keeping it secret NewYork Times,

10. Dezember. https://www.nytimes.com/interactive/2018/12/10/business/

location- data- privacy- apps.html. Accessed: 11. Mai 2020

Vinten-Johansen P (o. J.) e John snow archive and research companion.

https://johnsnow.matrix.msu.edu/index.php. Accessed: 1. Juni 2020

Visualising Advocacy (o. J.) Visualisation tools. Visualising Advocacy. https://visu-

alisingadvocacy.org/resources/visualisationtools.html. Accessed: 11. Mai 2020

H.-W. Eckert

105

Fair Play: What Counts inData Stories

Abstract It is through selection and interpretation that data gain their

meaning. is chapter is about the false certainties and deliberate manip-

ulations that inuence us in the interpretation of data and how we can

protect ourselves from them. e more power is attributed to data, the

more ethical questions gain relevance: the protection of privacy, the role

of algorithms in making decisions and the bias of machines come

into view.

Data is the new smoke: e thesis formulated at the beginning of this

book (see Sect. 1.2) aims at the fact that today we ascribe a large and

growing role to data in explaining the world. It should have become clear

that it is not the data themselves that contain this information, but we

attribute it to them. Only through selection and interpretation does the

data acquire meaning. To what do we direct our attention? is question

is more relevant than ever in view of the overabundance of possible infor-

mation. We do, after all, have a limited attention capacity. Every story

that is told and shared claims our attention and crowds out countless

other stories that are not told and shared. What narratives does the story

part of Springer Nature 2022

H.-W. Eckert, Storytelling with data, https://doi.org/10.1007/978-3-658-38555-2_5

106

being told dock onto? e more deeply a story is embedded in already

known and accepted narratives, the more likely it is to generate resonance

itself. Context is an important factor in its success. But beyond that, so is

its novelty value, that is, its deviation from familiar patterns. e task of

discovering meanings in the data and making sense of it is for the pythias

and the priests of the data oracles. And only when we believe the prophe-

cies and forecasts and act on them do they become eective.

Precisely because this process of meaning-making has so much inu-

ence on our actions, it is worth taking a look at misinterpretations and

manipulations. is chapter is about outlining which false certainties and

deliberate manipulations inuence us when interpreting data and how

we can protect ourselves from them. Finally, we look at voluntary initia-

tives and legal frameworks for storytelling with data.

5.1 False Certainties

andDeliberate Manipulation

One of the biggest sources of error in storytelling with data is that we

believe there is something like objective, incontrovertible truth in the

data. e more we believe this, the more powerful the inuence of the

stories we draw from the data. At the same time, we should keep in mind

that our perception works quite dierently. e human brain is not a

rational calculator. It loves drama and the emotions it evokes: e more

of it, the better it can remember the content. Stories with a clear downfall

spread well. is is how our ancestors were able to pass on vital experiences.

is was helpful in making quick decisions on the steppe: Fight or

ight. Sometimes fractions of a second were decisive in the hunt. At the

sight of a sabre-toothed tiger, there was no need for a sophisticated set of

subjunctives and possibilities. It was drama in its purest form. is pat-

tern is deeply inscribed in our brains, much like our preference for sugar

and fat, which provide our bodies with vital energy. is has ensured our

survival for the past tens of thousands of years. For living in our world

today, these patterns are not always helpful and are sometimes counter-

productive. Our cravings for sugar and fat fuel entire industries that

H.-W. Eckert

107

provide us with food that is more dangerous to our lives than all the wars

in the world (see Harari 2017, p.14). And our instinct to divide things

into good and evil, black and white, rich and poor, sometimes blocks the

view of dierentiations that give us valuable insights into our world and

form the basis for new stories and options for action.

Over the millennia of its development, our brain has been programmed

for survival. Rationality and reason are not among its vital functions, but

fast intuitive action is all the more important. us, in the course of evo-

lution, a kind of division of labour between two systems has developed in

our brain. e psychologist Daniel Kahneman has called these systems 1

and 2, or “fast thinking” and “slow thinking”. “Fast thinking” allows us

to act intuitively and emotionally. It works like a kind of autopilot.

Without pause, it makes judgments about distances, dangers or moods,

for example, and gives us the condence in everyday life that we are

always in control of the situation. Experienced experts can make precise

intuitive decisions. For example, a reman can sense the danger of an

impending explosion, a chess master knows the next appropriate move

without thinking. System 1 is receptive to simple judgments, clear polar-

izations, great drama. However, it is of no use in understanding more

complex relationships.

is is where “slow thinking” comes into play. is type of thinking, also

called system 2 by Kahneman, switches on when something complex or

unexpected arises in the stream of fast thinking. Slow thinking consumes

more energy, it is exhausting and has only limited capacities. Above all, how-

ever, it is characterized by its laziness: it usually lets system 1 take the lead

in interpreting sensory impressions, but always believes that it is in control.

According to Kahneman, the interaction of both systems determines

our thinking and our perception with all its false certainties and distor-

tions. For it is precisely slow thinking that lulls us into a sense of security

and makes us believe that we are in control of the situation, while fast

thinking has long since made the decision. In short, we make our deci-

sion emotionally, only to justify it rationally (Kahneman 2011,

pp.19–30). Applied to our topic, this means that we provide the rational

justication with the data. In many cases, however, we decided long

beforehand to believe the story and that is why we chose it in the

rst place.

5 Fair Play: What Counts in Data Stories

108

Storytelling can address both systems, the fast thinking with dramatic

eects, clear polarization and great fall, the slow thinking with irritations,

disruptions and breaks. It is in our hands to choose which story we tell.

In doing so, it is helpful to be aware of the traps we can fall into. ese

traps consist of the distortions of our perception that are due to the con-

struction of our wetware, i.e. our brain.

e Swedish physician Hans Rosling has devoted an entire book to

these distortions of our perception. In Factfulness, Rosling illustrates

these patterns of our thinking shaped by instincts, that is, what Kahneman

called quick thinking (Rosling 2018). He makes clear how to overcome

these narratives and retell stories. e technique is also called reframing:

familiar concepts are put into a new context and result in a dierent story.

A frequently recurring pattern is thinking in polarities. In many cases,

the division of our world into two parts prevents us from understanding

complex interrelationships. Rosling shows this in the distinction of the

world into rich and poor, into developing and industrialized countries,

into North and South. All these are variants of the same narrative that

divides the world into two and determines our view of politics, econom-

ics and society. Rosling recounts an experience in the 1990s when he was

discussing infant mortality in the world with his students and uncovered

this pattern: the “us” of the rich industrialized countries versus the “them”

of the developing countries. To the ght against this “mega misconcep-

tion” he had dedicated himself. Above all, Rosling’s approach makes one

thing clear: you don’t ght such a powerful narrative with individual

facts. To be truly convincing, you need a completely new narrative that

gives us a new perspective on things.

Rosling develops this narrative on the basis of the distribution of

wealth. For this purpose, he chooses per capita income and life expec-

tancy as indicators. His famous chart (Fig. 5.1) shows income (gross

national product per person and country) on the X axis and life expec-

tancy on the Y axis. e countries of the world are shown in circles, the

size of which represents the size of the population.

As might be expected, when the countries are arranged with the cur-

rent data, a diagonal line emerges that points to a correlation between

wealth and life expectancy. Or to put it more simply: the wealthier I am,

the longer I live.

H.-W. Eckert

109

Fig. 5.1 With this illustration, Hans Rosling shows that wealth in the world can-

not be divided between rich and poor, but is rather concentrated in the middle.

Interactive graphic at www.data- storyteller.de. (Source: Rosling, own

representation)

But this chart makes something else clear: there is no discernible polar-

ity of a distribution into rich and poor. For if there were, the picture

would look dierent, showing most or the largest circles (i.e. the coun-

tries with the largest populations) on the X-axis on the left (i.e. the

“poor”) side, while there would be a few rich countries on the right. But

there should be a gap in between if the picture of poor vs rich is correct.

But the picture shows a dierent distribution, namely a relatively strong

middle and a decrease towards the edges.

is leads Rosling to replace the dichotomy of “poor” vs. “rich” with

four development levels, which are also a simplication, but better reect

the distribution of wealth (Fig.5.2). Here, Level 1 (greatest poverty) and

Level 4 (greatest wealth) each have roughly equal numbers of people, at

one billion, while the bulk of humanity (5 billion) is distributed across

5 Fair Play: What Counts in Data Stories

110

Fig. 5.2 According to Hans Rosling, four income levels correspond better to the

distribution of wealth than the division into rich and poor. (Source: Rosling 2018)

the two middle levels.

Rosling thus replaces our polar pattern of think-

ing with a picture of four development levels in the form of a normal

distribution: the centre of gravity in the middle and signicantly less at

the two poles “poor” and “rich”.

e rich vs. poor thought pattern is based on a tendency to polarize,

which appeals to quick thinking. Polarization makes life easier by draw-

ing clear boundaries. Such a tendency of our brains to see clear patterns

can, of course, be used specically to emphasize dierences that are not

present in the data. Such overemphasis can be implemented by simple

means, such as cutting o a scale. e boundaries from focusing to

manipulation are blurred here. e cropping forms an enlargement of

the actual image. To illustrate this, here are two representations of values

in a bar chart– one with a heavily cropped Y-axis and dierent colour

markings and one with the complete scale from 1 to 100 and the same

colours (Fig.5.3).

e truncation of the Y-axis highlights dierences much more than

they actually are in the percentages, namely 62% versus 54%. To avoid

such manipulations, Datawrapper, for example, always lets the scale of

the Y-axis start at 0. Only at the top is there the possibility to limit it.

Rosling (2018, p.33). e graphic on the endpaper of the book is available as an interactive rep-

resentation at www.gapminder.org. An implementation of the graph in the R programming lan-

guage has been developed by Keith McNulty: McNulty (2019).

H.-W. Eckert

111

Fig. 5.3 Trimming the Y-axis ampliﬁes the differences. (Source: Own

representation)

5 Fair Play: What Counts in Data Stories

112

Excel, on the other hand, presents a cropped view with the same data and

can only be expanded to a full scale by intervening in the axis options.

However, there is a lot of potential for misinterpretation or deliberate

manipulation not only in the cutting of the scales, but already in the

preparation of the data. Hans Rosling shows this with the example of the

use of averages.

Averages can emphasize dierences that are not present in this form in

the raw data. Rosling explains this with the example of math skills of

women and men. Multi-year comparisons reveal dierences in math

skills between women and men based on SAT (Scholastic Assessment

Tests) tests in the US.Men perform slightly better than women, although

the gap has been steadily narrowing since the 1980s. However, when

looking at the within-year (in this case 2016) distribution rather than the

yearly averages for math scores, a picture of two hump-shaped distribu-

tions emerges that are slightly skewed. e majority of women are just as

procient in math as men. Polarities built up by means of averages, his

message goes, are therefore not always a source of insight. ey are espe-

cially not so when there is a normal distribution between the two poles,

as in this case (Rosling 2018, pp.40–41).

Deliberate manipulation or an invitation to misinterpretation can also

occur through the use of colors. e meaning of colors is deeply anchored

in our culture: we see red, make blue or are green with envy. e use of

these colors in a graphic cannot be separated from the context of color

meanings. It is debatable whether colors have a universal meaning, or

whether their roots lie in our respective cultures. Probably both aspects

overlap (see the empirical study by Jonauskaite etal. 2020). For example,

the meanings of colors dier by culture. In China, red is the colour of

life, which is also said to bring good luck. In our Western European cul-

tural circle, on the other hand, red stands for danger or aggression, among

other things. It should therefore always be clear in which cultural context

a graphic is used.

is also applies to color combinations and contrasts: For example, the

color pair red/green stands for stop/go or negative/positive. In graphics,

this is used in analogy to the trac light system, for example, for rising

and falling directions in price trends or dos and don’ts in recommenda-

tions for action. e use of these colours as accents will therefore always

H.-W. Eckert

113

have such a connotation. e designer should consider whether this is

desired or possibly misleading.

e contrast red/blue has been given a further context by the climate

debate: in this case red stands for warm, blue for cold. is has given rise

to a new form of graphic representation: Series of vertical stripes repre-

senting temperature corridors. e climate scientist Ed Hawkins from

the National Centre for Atmospheric Science (NCAS) at the University

of Reading has made this form of graphic known with his Warming

Stripes (Fig.5.4): a colour bar stands for a specic average temperature

segment. In this way, long-term changes can be made visible to the eye.

As simple as the representation is, as complex are the calculations behind

it, because each colour nuance stands for a certain temperature corridor

in a previously dened period of time. Yet the result is so captivatingly

intuitive and has been so successful that the graphic format, which was

rst introduced in 2018, is now available for a large number of countries

and time periods and is still being further developed (cf. Müller-

Jung 2019).

In June 2019, the endeavor resulted in the “Show your stripes” initia-

tive, which now features graphics for nearly every country from 1901 to

2018 at https://showyourstripes.info/.

Fig. 5.4 Making long-term changes visible: Climate scientist Ed Hawkins’

Warming Stripes document climate change. (Source: Wikimedia Commons https://

commons.wikimedia.org/wiki/File:20181204_Warming_stripes_(global,_

WMO,_1850- 2018)_- _Climate_Lab_Book_(Ed_Hawkins).png)

5 Fair Play: What Counts in Data Stories

114

e language of this representation is intuitive: blue stands for cooler

temperatures, red for higher. As temperatures rise, it follows that the bar

on the right-hand side of the image contains more and more red. is is

a deliberately chosen color strategy that immediately makes the underly-

ing narrative clear to any viewer: it’s getting warmer.

In the meantime, climate change deniers have also taken up this

colouring and are trying to counter this reading with a counter-strategy.

Birgit Schneider, a professor of media ecology in Potsdam, is investigat-

ing the images of climate change and is working on a study for which she

collects images of scientists and climate change deniers and analyses the

respective colour strategy. She can prove that climate change deniers rely

primarily on the color blue. When they adopt red, it is to show how the

opponents, i.e. the alleged ‘simulators’, have been deceived

(Schneider 2020).

e more we know about these processing techniques and our brain’s

secret weaknesses, the better: when we know them, we not only read

images and stories dierently, but we also nd new ways to nd and tell

stories. is also applies to the following heuristics:

• Everything used to be better: many narratives are based on this pat-

tern. e dying of the forest, the distribution of wealth, the rise of

crime. It’s related to idealizing our memories and overweighting pres-

ent woes in comparison. “In human memory, the good is often stron-

ger than the bad,” psychologists Constantine Sedikides and John

Skowronski write about this. One reason, they suggest, is that good

memories have stronger aective content: Feelings associated with rosy

nostalgia have more emotional power than feelings with negative

thought content, according to the paper (Sedikides and Skowronski

2020). In Factfulness, Rosling goes against this reading and narrates

the development of humanity as a process of steady improvement–

less poverty, falling crime rates, more education, rising prosperity and

less inequality.

• Extrapolation – the extension of the straight line: An obvious but

often incorrect assumption is that an observed development continues

to progress uniformly: is is not true for the outbreak of an epidemic,

nor for the price trend of a stock. We nd it dicult to predict expo-

H.-W. Eckert

115

nential growth or even the reversal of directions. For many years, the

development of globalization also seemed to know only one direction:

towards increasingly interconnected international value and sales

chains. e nancial crisis, the strengthening of China and the rise of

nationalist movements in Europe and the USA have put the brakes on

this development. e outbreak of the Corona pandemic at the latest

showed how fragile this order is.

But can such developments actually be modelled using mathematical

methods and thus forecasts developed for the further course? Curve t-

ting describes the procedure for describing a curve from data using math-

ematical functions– for example by interpolation or regression analysis.

Most visualization programs, such as R, Matlab, and tools based on them,

provide a range of curve ts for dierent scenarios. ey lull us into the

certainty that mathematical functions can predict developments.

• Simple numbers: A number without context means nothing. Are 805

million tonnes of greenhouse gases in Germany in 2019 a lot or a lit-

tle? Only in relation does the number gain meaning: for example, in

comparison to the previous year or over a longer period of time. Or in

comparison to other economies, to the size of the population, etc.?

e same applies to sales, prots of companies, to the evaluation of

website statistics and engagement rates of social media posts: e com-

parison raises the anchor and determines the rest of the narrative.

• Generalization: Generalizing helps us to orient ourselves in everyday

life and is an essential characteristic of quick thinking. We use it to

transfer familiar patterns to new situations. However, it obscures the

view of other perspectives. erefore, it is sometimes useful to question

the categories into which we divide our world: be it consumer habits,

educational level or nancial needs. It is worth looking at dierences

within large groups, as well as similarities and dierences between

groups. A critical look also applies to orders of magnitude: For exam-

ple, is a majority made up of 51% or 99%? is is a big dierence

when it comes to assessing how relevant this group is compared to

other groups.

5 Fair Play: What Counts in Data Stories

116

• Conrmation bias: People tend to trust information that supports

their attitudes. Information is selected, ascertained and interpreted in

a way that fulls or conrms one’s own expectations. is is particu-

larly the case in complex and emotional debates, such as climate

change or elections. But conrmation bias also plays an important role

when deciding on a product, because this usually consists of a bundle

of benets and cannot simply be compared with another product from

a competitor. When analyzing and interpreting data, it is important to

be clear about one’s hypotheses as early as the search stage.

In com-

munication, however, conrmation bias is used very deliberately.

Brand strategies come into play here: they oer the reduction in com-

plexity that System 1 “wants”. Brands build trust by conrming and

supporting us in our attitudes. ey prevent system 2 from interven-

ing and critically questioning how much sugar is in the spread or what

is in the ne print of the insurance conditions.

• Standstill: We have no sense of slow change. So some changes extend

over such a long period of time that they seem like stagnation to our

perception. Our instinct then tells us that these things are unchange-

able. For example, ideas about nations, cultures, and economies. Much

data is available for longer periods of time, but media and communica-

tors need many occasions to produce news. at’s where shorter time

periods help. Best-seller lists work like this: Weekly updates produce

news, whereas if viewed over centuries, the Bible and Koran would

always be ahead. How boring. But when assessing the relevance of cur-

rent topics such as globalisation, industrialisation or digital transfor-

mation, it is worth looking at longer periods of time to see what

momentum the terms have and how relevant they will be in the future.

As in the example of the term eld of articial intelligence (see Sect.

3.3.3): Are mentions and search entries rising to new heights, is the

curve attening out or is it already sinking? Are there new topics from

the eld whose attractiveness is increasing or is the whole technology

already heading for the next AI winter? And of course, one should also

en a positive test strategy can help to limit the set of all possible test parameters to a plausible

and practicably veriable selection, cf. for example Pohl (2004).

H.-W. Eckert

117

be aware that the viral course with a hump-shaped development is

again only an assumption from the pattern of an infection course.

• Simple is attractive: Simple solutions appeal to our System 1. Finally

getting clarity about a connection triggers feelings of happiness. e

world becomes explainable and causes can be traced to an eect. e

idea of the free market must declare all responses of government inter-

vention to be false. It simply eliminates all arguments from the oppos-

ing side. If we have only one tool, we have a solution for everything.

• e Availability Trap: Anyone who has ever read statistics about bur-

glaries in their neighborhood tends to overestimate the frequency of

such incidents. We estimate the likelihood of something happening

based on the examples that come to mind. It is a shortcut to estimating

the situation. Anyone can fall into an availability trap by looking at

numbers and data and using existing material. For example, does the

engagement rate of my social media posts allow me to make a solid

statement about the success of my content strategy, or do other factors

need to be included? Such as: the target groups reached, their purchas-

ing power, the conversion rate, etc.?

• Survivorship Bias: Successes are systematically overestimated because

they are more visible than defeats. e term goes back to Allied engi-

neers in the Second World War. ey examined the bullet holes of

ghter planes. ese were mainly in the area of the wings, tail units

and in the middle of the fuselage. erefore, they initially reinforced

these areas, but this did not improve the rate of return. From this the

insight was formed that just the places must be strengthened, which

were not aected with the returning airplanes: above all pilot pulpit

and engines. Because these weak points ensured that the other aircraft

did not return.

e examples show that even if a part of our thinking prefers simple and

quick truths, we do not have to accept them immediately. A close look at

the data can protect us from misinterpretation and help us discover new

narratives. It helps to check whether the data really support our interpre-

tation or whether we have fallen for false certainties and the distortions of

our own perception.

5 Fair Play: What Counts in Data Stories

118

5.2 Machine Bias

e more machines take over the processing of numbers, texts and

images, the more we also hand over part of the power of interpretation to

them. However, a machine is only as good as the data it works with.

Among computer scientists there is the saying: “Garbage in– Garbage

out”. But the garbage we produce with data can also be toxic. Jim Balsillie,

founder of Blackberry’s manufacturer Research in Motion (RIM), sums

up this danger with a drastic comparison: “Data is not the new oil– it’s

the new plutonium. Amazingly powerful, dangerous when it spreads, dif-

cult to clean up and with serious consequences when improperly used.”

For Balsillie, this danger is particularly great when the data ends up in the

hands of large technology corporations: Corporations like Facebook and

Google would even replace the press as the fourth power in the state and

are a danger to liberal democracy.

Even if one might consider this view

to be exaggerated, it is worth taking a look at the toxic side eects of our

data-centric economy.

When processing data, there are two main sources of error: the selec-

tion and the processing of the data. e biases that arise from these errors

are collectively known as machine bias or algorithmic bias. One of the

originators of the critical view of computer science was Joseph

Weizenbaum, who analyzed the methodological basis of bias in his 1976

book Computer Power and Human Reason (Weizenbaum 1976).

Prejudices can already be contained in the selection of data. Many

methods of data processing are also based on interpretations, sometimes

and possibly very often also on premises that are not highlighted clearly

enough or are even erroneous. As soon as machines automate the process-

ing of large amounts of data, this eect is potentiated. is is why a criti-

cal look at the sources and their processing is so important. In the case of

data that a company collects from its own sources, it is still easier to trace

the origin and therefore also its meaningfulness. With data sourced from

Balsillie before the International Grand Committee on Big Data, Privacy and Democracy in

Ottawa on May 18, 2018.

“Technology is disrupting governance and if left unchecked could render liberal democracy obso-

lete. By displacing the print and broadcast media in inuencing public opinion, technology is

becoming the new Fourth Estate.” ibid.

H.-W. Eckert

119

third parties, this is much more complex. e more Big Data consists of

recycled and highly aggregated data, the greater the risk of such errors.

Data selection: A common argument for data mining with algorithms

is that human bias and prejudice play no role in decision making. Yet

data mining, if used without reection, can reproduce the very patterns

of discrimination, biases of decision makers or even of parts of society.

Decisions, for one, can be made on the basis of inaccurate, incomplete,

or even non-representative data. Even high-quality data can contain sta-

tistical biases if they do not accurately represent the proportions of cer-

tain groups. For example, to detect road damage, the city of Boston has

developed an app that uses a smartphone’s accelerometer. If someone

drives through a pothole, the app records it. However, as ecient as this

approach is, there is a risk that the recording will disadvantage poorer

parts of the city because fewer people there have smartphones with cor-

responding sensors.

Second, the data itself may already contain biases. Many intelligent

systems today learn from data that comes from the internet and social

media channels. Biases contained therein are then reproduced in the pro-

cessing by the algorithms or AI.A well-known case is the scandal Google

produced with an earlier version of its facial recognition software, Google

Photos. In 2015, the software classied images of dark-skinned people as

“gorillas.” Google was shocked and improved the system. But Google is

not alone in the bias. Other systems also recognize white faces better than

anyone else. For example, facial recognition software from Microsoft,

IBM, and the Chinese company Face++ is also particularly good at recog-

nizing white males (Kaltheuner and Obermüller 2018).

Meanwhile, language has overtaken image recognition as the branch of

AI with the greatest appetite for data and computing power. Again, there

are numerous biases reproduced from the data used and incorporated

“Nowadays, we don’t have direct data. We are recycling data and using proxy data– things like

how you click on websites, what you purchase, what you say on Twitter, who your friends are on

Facebook– to infer things that we are interested in. e promise of big data is that we will be able

to use all this proxy information to determine with increasing accuracy the things that we care

about.” Interview with Cathy O’Neill– Burack (2017).

See, for example, the detailed account of errors in data selection and preparation in Barocas and

Selbst (2016), here p.685.

5 Fair Play: What Counts in Data Stories

120

into further processing. Open AI, a non-prot organization that addresses

the “existential threat of AI,” also develops its own open-source software.

One of their most powerful tools is Generative Pre-Trained Transformer

3 (GPT-3), a software that writes text. e system is capable of producing

impressive literature of a wide variety of genres– short stories, sketches

and even poems– that are sometimes barely distinguishable from texts

written by humans.

e system is based on a model that calculates and creates words, sen-

tences and paragraphs based on statistical predictions. And therein lies

the problem: statistical methods are no substitute for a coherent under-

standing of the world. GPT-3, like all other AI systems, has no internal

model of a world, its values, or its narratives. erefore, it cannot ulti-

mately provide reasoning that requires such a model.

Rather, this system also reproduces the prejudices with which it has

been fed. And since it obtains the necessary quantities of texts from the

Internet and social media channels, its “view” of the world also comes

from these sources. Because of its statistical approach, texts containing

terms such as “black,” “Jew,” “woman,” and “gay” are often associated

with prejudices that include racism, anti-Semitism, misogyny, and

homophobia. Given the amount of training data required for such a lan-

guage engine, any attempt to establish adequate quality control is doomed

to failure.

e processing of the data: Biases do not only arise from the data base.

Further processing via algorithms or AI also make it necessary to ask

questions about the ethics, transparency, and accountability of the sys-

tems used. Pro Publica’s editorial board investigated software widely used

in the US criminal justice system to determine whether a convict should

receive parole. Because the software company was unwilling to disclose

the details of how the software worked, Pro Publica used convict data to

do a kind of “reverse engineering” of the software system. It found that

this software reproduced racism. e software rated the likelihood of

people of color to recidivate signicantly higher than for whites (Angwin

etal. 2016).

Cathy O’Neill has written a book about the biases that US govern-

ment agencies and corporations use to make decisions based on data,

thereby cementing prejudices and disadvantaging social groups. In

H.-W. Eckert

121

Weapons of Math Destruction, she shows how these procedures are used

in the labor market, education, banking and insurance, policing and jus-

tice, and, of course, advertising (O’Neil 2017). All procedures are based

on large amounts of data and the belief of authorities and companies that

by modelling this data they can not only automate decisions but also

eliminate human inuences. But the data collected also contains biases

on the basis of which decisions are made about loans, access to education,

or even arrests. us, the exact opposite happens: namely, the software

draws its conclusions based on the data put there by humans and their

human-programmed processing, which not only contains errors, but also

reproduces them and thus further reinforces them.

Reminiscent of the movie Minority Report is the controversial PredPol

system that the LA Police Department implemented in 2011, which was

only suspended after massive criticism under the pretext of the Corona

crisis. PredPol is crime-ghting software that uses historical data to calcu-

late the likelihood of future crimes. e system remains in use in many

police departments around the world. According to the manufacturer,

only three data points are needed for prediction: Type of crime, location

and time. No demographic or ethnic data would be used in the analysis

(Lobe 2019, p.207). But they are not needed, because they are already

included in the three data points through correlations.

e reason why US-American examples are listed here is that auto-

mated decision-making processes in politics, administration and business

are particularly widespread there and correspondingly many examples of

such distortions are also documented. e questions that arise from this

are also relevant for communication and marketing managers in German

companies:

• Anyone using a large collection of sources from the internet and social

media for their analysis should be aware that these sources contain bias.

• Many tools for automated speech, image, and text recognition and

prediction using AI may contain processing errors that arose from the

test data used.

• When analyzing their own data (such as CRM, website, shop, sales,

service), communication and marketing managers should always ask

themselves how representative these results are and where there may be

5 Fair Play: What Counts in Data Stories

122

blind spots, for example because certain user groups are underrepre-

sented. is can lead to certain target groups not even coming into

view because they have hardly come into contact with the company so

far and have therefore not left any relevant data.

at’s why it’s important that today’s pythias and priests in companies

understand how the data they work with came about and what the prem-

ises are. ey need to be comfortable with technologies that allow them

to tap into and interpret their own data sources. ey need knowledge of

sources, market knowledge of common tools, methodological skills in

analyzing and preparing data. And they should be familiar with common

heuristics.

5.3 Ethical Issues oftheData Oracle

Dealing with the distortions of our perception quickly leads into the

realm of ethics. For it is generally about questions that guide our actions:

What goals are we pursuing? What reasons do we have for doing so and

what means do we use to achieve them? Precisely because data has such a

great inuence on our world through digital technologies, it is necessary

to consider developments also in terms of those problems and conicts

that are triggered by them. e discussion about ethical issues in dealing

with information technology is as old as the discipline itself. Norbert

Wiener, one of the founding fathers of articial intelligence, already

addressed the ethical dimensions of his discipline in his standard work

Cybernetics (1948). Even during the war, Wiener foresaw enormous

social implications of technology. In this context, he spoke of a second

industrial revolution with enormous potential for good and evil. It was

clear to Wiener that this would bring with it a host of ethical challenges

and opportunities. In his subsequent works, he explored a range of ethi-

cal issues that computers and information technology would raise. He

identied a variety of issues that continue to inform data ethics today:

ese included implications for security, for the labor market, for the

responsibilities of computer specialists, for information networks and

globalization, for virtual communities, telecommuting, transhumanism,

H.-W. Eckert

123

and robot ethics (Wiener 1948, pp.169–180; Bynum 2018). However,

the topic did not really go viral until the beginning of the new millen-

nium, when the impact of technologies was clearly felt with the advent of

cloud technology and the enormous availability of data (Big Data)

(Fig.5.5).

e more data is traded as a commodity of the twenty-rst century

and processed by machines, the more important it becomes for data sub-

jects to know what happens to their data. Questions about informational

self-determination are more relevant today than ever before. Business

models worth billions are based on the analysis of user data, the Internet

of ings is increasingly nding its way into our everyday lives in the

form of wearables or smart homes, gigantic amounts of data are managed

in data warehouses and we can all pay for supposedly free services by giv-

ing up our data.

Fig. 5.5 For a long time, the ethical handling of data was a topic for specialists.

The topic only gained greater relevance at the beginning of the new millennium,

when with the advent of cloud technology and the enormous availability of data,

the effects of the technologies became clearly noticeable. Interactive graphic at

www.data- storyteller.de. (Source: Google Ngram Corpus English 2019)

5 Fair Play: What Counts in Data Stories

124

With its growing importance, data ethics has also left the eld of spe-

cialists and gained general relevance. At the latest with the revelation of

the activities of the US intelligence services by Edward Snowden in 2013,

governments have recognised the importance of the global cloud infra-

structures of the data economy. Even if the focus here was not primarily

on the question of personal rights, but rather on the defensibility of

states, this has initiated a rethink. e fact that the preoccupation with

the ethical consequences of technological developments has arrived in the

industry is also documented by the market research company Gartner,

which has selected “digital ethics” as the top technology trend for 2019.

And things are also moving at the political level: With the General

Data Protection Regulation (GDPR), which came into force in May

2018, Europe has taken a stand for the protection of personal rights

and thus set global standards. e German government has launched

an AI Enquiry Commission as well as a Data Ethics Commission. In

October 2019, the Data Ethics Commission presented its report, which

contains 75 recommendations on the handling of data and algorithmic

systems, including articial intelligence (Report of the Data Ethics

Commission 2019).

5.3.1 The Protection ofPrivacy

Since May 2018 at the latest, the next page on the Internet is no longer

one but two clicks away. In the meantime, we have long since become

accustomed to rst conrming the privacy notices of each website when

surng the internet before moving further on the page. is is the most

visible, but by no means the only consequence of the introduction of the

General Data Protection Regulation (GDPR).

e regulation, which is valid throughout the EU, is primarily con-

cerned with the personal and fundamental right of informational self-

determination. It is derived, for example, from the general right of

personality described in the German Basic Law (Article 2 (1) GG). is

refers to the right of every individual to be able to decide for himself or

herself on the disclosure and use of his or her personal data. is is in

tension with economic interests, which are a major driver of value cre-

ation with the collection and further processing of data. us, the GDPR

H.-W. Eckert

125

tries to balance the protection of personal data within the European

Union on the one hand and the guarantee of a free movement of data

within the European Single Market on the other hand.

e GDPR is polarizing: While some are of the opinion that the rules

overshoot the mark, others consider data protection to be a good thing.

However, criticism of the regulation is not so much ignited by the protec-

tion of personal data, but by the poor implementation of the directive.

Marco Zingler from the German Digital Economy Association (BVDW)

criticised: “When it came into force in May 2018, the GDPR put a

damper not only on the digital industry, but on the entire economy in

Germany and Europe. Not because of overly strict data protection regu-

lations. e most serious problem is the legal uncertainty caused by con-

tradictory and unclear formulations of the regulation” (Eichsteller and

Seitz 2019, p.13). is uncertainty is also reected in the results of the

Digital Dialog Insights 2019 survey.

While less than a third of the study participants have doubts about the

ROI of data-driven communication, concerns about the permissibility of

data collection and uncertainty caused by legal developments are the main

hurdles to the expansion of data-driven marketing communication (Fig.5.6).

Fig. 5.6 Doubts about the legitimacy of data collection and uncertainty caused

by legal developments are the main hurdles to the expansion of data-driven mar-

keting communication. In contrast, only a few question the ROI of data-driven

communication. (Source: Eichsteller and Seitz 2019)

5 Fair Play: What Counts in Data Stories

126

Criticism has also been levelled at the high costs and the not always

practical requirements that implementation of the requirements has

entailed, especially for smaller companies. However, it is an important

milestone for securing the rights to one’s own data and a step in the right

direction– namely positioning the EU as a pioneer in the protection of

personal rights in the digital age. e GDPR is based on “privacy by

default”: primarily, strict protection of personal data applies. Only with

the consent of the data subject may these be weakened. e General Data

Protection Regulation is generally aimed at the handling of data– online

and oine.

e GDPR is shaped by the idea of a single European market for data.

us, data may only leave the EU if companies have adequate security

measures in place or if the destination country has an “adequate level of

protection”. Such a level of protection is no longer provided by the

EU-US data protection agreement, as ruled by Europe’s highest court,

the ECJ.In their ruling of 16 July 2020, the Luxembourg judges found

that the so-called “Privacy Shield” is invalid. When transferring data of

European consumers to a third country, a level of protection at the level

of the GDPR had to be maintained. e US legislation does not provide

a basis for this. ey demand that the transfer of personal data must com-

ply with the level of protection required by the EU.

But a regulation does not create a market. e EU is therefore develop-

ing the next steps for a single market for data, along the lines of the single

market for goods. e EU website says: “e European Data Strategy

aims to put the EU at the forefront of a data-driven society. By creating a

single market for data, it will be possible to share it within the EU and

across sectors for the benet of businesses, researchers and public admin-

istrations.” (European Commission 2020). A European cloud solution is

also intended to contribute to this: with GAIA-X, representatives from

politics, business and science are developing a proposal for the design of

a data infrastructure for Europe. e aim is to create a secure and net-

worked data infrastructure that makes Europe independent of non-

European cloud providers and promotes innovation.

e question

See the website of the Federal Ministry for Economic Aairs and Energy https://www.bmwi.de/

Redaktion/DE/Dossier/gaia-x.html

H.-W. Eckert

127

remains whether the EU will succeed in standing up to the de facto dom-

inance of US and Chinese cloud providers.

After all, when it comes to protecting personal data and limiting the

ow of data, the GDPR has found many imitators. India keeps payment

information within its own country and may soon require that certain

types of personal data not leave the country. Russia requires that data be

processed and stored on servers within its territory. China blocks most

international data ows. Most importantly, California has enacted the

California Consumer Privacy Act (CCPA), which went into eect in

January 2020, a regulation that rivals the European model in importance

and scope (Economist 2019, 2020).

5.3.2 Algorithmic Accountability

In addition to collecting data, companies are also increasingly in the pub-

lic eye when it comes to dealing with the machine processing of data. e

question of responsibility for machine results and their use in public

authorities, politics and companies is gaining social relevance. e com-

munications researcher and computer scientist Nicholas Diakopoulus

has coined the term “algorithmic accountability” for this. His report on

the study of algorithms as black boxes appeared at the beginning of 2014,

and since then the term has had a rm place in the public debate (Fig.5.5).

In it, Diakopoulos describes a new task for journalists to understand

software systems as objects of reporting. In addition to the programming

code, the data sets with which these systems are trained must also be

examined. Otherwise, their modes of operation cannot be understood

(Matzat 2017).

e FAIR Data Initiative focuses on the processing of data by machines.

It emerged from an association of data scientists who developed the FAIR

principles in 2016. FAIR stands for

• Findable

• Accessible

• Interoperable

• Reusable

5 Fair Play: What Counts in Data Stories

128

Germany, France and the Netherlands have taken up the initiative and

created GO FAIR, a body to support it (Wilkinson etal. 2016).

e industry association for the digital economy, Bitkom, has made

recommendations for the responsible use of AI and automated decision-

making and advocates that companies develop internal guidelines on the

use of algorithms. “is can take the form of corporate ethics, for exam-

ple. Impact assessments that are developed in relation to the use of algo-

rithms should be incorporated into the development of algorithms. An

ecient and at the same time agile process must be anchored that regu-

larly adapts these guidelines to new technologies and the issues that arise

with them.” (Bitkom 2018).

e European Union has taken up the issue and is in the process of

dening the framework. In the 2019 study “A governance framework for

algorithmic accountability and transparency”, it outlines four policy areas

for action:

1. Awareness raising: Education, watchdogs and whistleblowers,

2. Accountability in the use of algorithmic decision making in the pub-

lic sector,

3. Regulatory oversight and legal liability in the private sector, and

4. e global dimension of algorithmic governance (European

Parliament 2019).

5.3.3 Voluntary Commitments

e most important step in achieving algorithmic accountability is the

willingness of companies to take legal and ethical responsibility for it.

First and foremost, this includes an understanding that there are sources

of error and bias in this area. Awareness of this, however, has room to

grow, as a study by consultant Deloitte shows. In its 2020 AI study,

respondents see algorithm bias as much less relevant than, say, security

concerns and transparency. It is little consolation that the sensitivity of

the respondents from Germany is two percentage points higher than

worldwide– especially since the respondents were experts in companies

that already use AI technologies (Fig.5.7).

H.-W. Eckert

129

Fig. 5.7 Security concerns and lack of transparency dominate the evaluation of

AI, while a possible bias of the algorithms is assessed as less relevant, Germany

n=201, Global n=2737. (Source: Deloitte 2020)

One of the rst corporate initiatives in this area was the Partnership on

AI initiative (https://www.partnershiponai.org/) by the major US tech

companies. Amazon, Alphabet, Facebook, IBM and Microsoft announced

it in September 2016. It has since been joined by more than 100 part-

ners. “In support of our mission to benet people and society, the

Articial Intelligence Partnership conducts research, organizes discus-

sions, shares insights, provides thought leadership, consults with relevant

third parties, answers questions from the public and media, and creates

educational materials that advance understanding of AI technologies,

including machine perception, learning, and automated thinking.” Right

after safety-critical AI (especially in health and transportation), Machine

Bias is addressed here as one of six topic areas: fair, transparent, and

accountable AI– is is about uncovering hidden assumptions and biases

in data, especially in the disciplines of biomedicine, healthcare, security,

criminal justice, education, and sustainability.

Salesforce, a US cloud software provider and member of the initiative,

created a dedicated position within the company for the ethical use of

5 Fair Play: What Counts in Data Stories

130

data in August 2018. Kathy Baxter is the architect for the ethical applica-

tion of articial intelligence (AI). To do this, Baxter works with Salesforce’s

research team, which develops the models for the company’s AI applica-

tion, Einstein. Here are the questions:

• Is there enough training data available?

• What are the possible reorganizations that we are shaping with

the model?

• And how do we mitigate that?

Subsequently, the product teams consider how to deliver a responsibly

usable product to the customer. Particularly sensitive information such as

a person’s age, ethnicity, and gender cannot be used to make decisions in

regulated industries (health insurance and nancial services) in the

United States. But even if these elds are disabled, it may be possible to

draw relatively accurate conclusions about these details through correla-

tions with, say, place of residence. A module in the Salesforce software

therefore checks precisely these correlations between the elds and identi-

es potential problems (Kutsche 2019).

In Germany, the debate on voluntary commitments by companies is

taking place under the term Corporate Digital Responsibility (CDR).

e term ties in with corporate governance and corporate responsibility

structures that are already established in many German companies. Since

2015, various companies have been developing their understanding of a

responsible approach to digitalisation and data. e paths and approaches

are as diverse as the companies themselves.

Companies from the retail, nancial services, IT, media and telecom-

munications sectors have emerged as the main drivers of this topic. ese

either have a particularly large amount of data or particularly sensitive

data, or play a role in the further processing of data or the development

of digital products. Depending on the market and the company’s busi-

ness model, the focus is either on the handling of personal data in general

or on the algorithmic processing of data in particular.

In 2018, the Federal Ministry of Justice, under the leadership of

Katarina Barley, launched a CDR initiative in which representatives of

Telekom, Miele, Otto Group, Telefónica Deutschland as well as DIE

H.-W. Eckert

131

ZEIT participated and to map out the framework conditions of corpo-

rate digital responsibility (cf. the press release of the Federal Ministry of

Justice “Responsibly shaping digitalisation” of 9 October 2018).

ese considerations are based on the common understanding that

sustainable development begins where legal requirements end. As a vol-

untary commitment by companies with a (partly) digital business model,

CDR addresses the side eects and risks of digitalisation that the com-

munity and individuals currently have to deal with, including the

following

• the “unethical” use of customer data and corruption of digital

self-determination,

• a non-transparent digital world whose rules cannot be controlled by

the community, e.g. the functioning of articial intelligence

In other words, it is equally about exploiting opportunities and averting

risks. e conviction behind this: Digital ethics, along with technological

competence, is part of the foundation of innovation and determines the

economic success of a company. e German Digital Economy

Association (BDVW) has taken up the topic and developed a framework

to support companies in the strategy and implementation of CDR (see

https://www.bvdw.org/der- bvdw/gremien/corporate- digital-

responsibility/cdr- building- bloxx/).

5.3.4 Data Literacy

But how can such a large topic be implemented in practice? Ethical stan-

dards can only be established if there is a broad discussion and corre-

sponding competence in wide circles of society. What is required,

therefore, is a responsible approach to data. Just as reading, arithmetic

and writing are among our basic skills, in the information age we need

skills for dealing with data. e term data literacy, best translated as data

competence, is used to describe the abilities to deal with data appropri-

ately, to interpret it and to present the results. “Data Literacy is much

more than a broad and deep detailed knowledge of constantly changing

5 Fair Play: What Counts in Data Stories

132

methods and technologies. Rather, the dimension of data ethics, motiva-

tion and value attitude plays a central role in being able to deal with data

successfully and condently in the future.” is is what Katharina

Schüller, Paulina Busch and Carina Hindinger write in their paper Future

Skills: a Framework for Data Literacy.

is requires skills from several disciplines– mathematics, statistics and

programming. In their interplay, they help us achieve the necessary com-

petence in dealing with data. us, data literacy is considered a key com-

petence of the twenty-rst century, essential for our society and working

world. Not at all a specialized discipline of computer scientists and statisti-

cians, but a cross-disciplinary skill for students of all disciplines, the

authors urge. “e process of knowledge creation involves several steps:

A. Establish data culture

B. Provide data

C. Evaluate data

D. Interpret results

E. Interpret data

F. Deduce action

In order to systematically create knowledge or value from data, the ability

to deal with data in a planned manner and to be able to consciously use

and question it in the respective context will therefore be of decisive

importance in all sectors and disciplines in the future.” (Schüller etal.

2019, p.10).

e communications disciplines have not yet played an active role in

establishing ethical data standards and building data literacy. e topic

has also not yet found its way into the standards of the industry associations.

Neither the Press Council, the Association of Professional Journalists, the

German Advertising Council nor the German Council for Public

Relations deal with data.

In the sub-discipline of data journalism, on the

See Press Code of the Press Council https://www.presserat.de/pressekodex.html, Code of Ethics of

the German Association of Professional Journalists DFJV https://www.dv.de/ueber-uns/ethik-

kodex, Guidelines of the German Advertising Council https://www.werberat.de/content/leitfaden-

zum- werbekodex-des-deutschen-werberats and German Communication Code of the DRPR

https://drpr-online.de/deutscher-kommunikationskodex/

H.-W. Eckert

133

other hand, there is no code, but there is a great sensitivity towards the

collection and processing of data. Here, people are very well aware of the

dangers. Behind this is the conviction that visualizations are only trusted

if the process of their creation is as comprehensible as possible. In addi-

tion to citing sources, describing the research, selecting and preparing the

data, access to the raw data also plays an important role in transparency

(Greussing 2019). It is therefore most likely that important impulses for

the communication disciplines and their handling of data can be expected

from this side.

5.4 Less Is More: AnOpportunity

forStorytelling

e collection of personal data is to be further restricted by the European

ePrivacy Regulation (ePVO). Even though it will not come into force

until 2022 at the earliest, the course has long been set for a world without

cookies from third-party providers (so-called third party cookies). e

interpretation of the GDPR by courts and the default settings of browser

providers have now largely anticipated the eects of the regulation.

e regulation aims to ensure that the analysis of user data through

cookies is only permitted if users have given their express consent before-

hand. Why does this have such major consequences? In a nutshell: is

will deprive the advertising industry and advertising companies of what

is currently their most important lever in the collection and personalised

targeting of advertising.

If you do not click on “accept all” when entering a website, you will get

an overview that looks something like Fig.5.8. e user can decide which

categories of cookies he accepts. Information from necessary, performance-

oriented and functional cookies remain with the website operator, the

cookies marked with advertising are the so-called third-party cookies,

which pass on information to advertising partners and their partners.

Since third-party cookies can be used to track users’ surng behaviour

across many websites and the collected data can also be resold, they are in

the focus of criticism. On this point, the ePrivacy Regulation is intended

to further specify the general “Privacy by Default” requirement of the

5 Fair Play: What Counts in Data Stories

134

Fig. 5.8 The user can decide to whom his data may be passed on. (Source: Own

representation)

GDPR and apply it to digital channels.

is stance is also reected in

two rulings: in October 2019, the European Court of Justice (ECJ) ruled

that users must actively consent to the setting of these cookies. In May

2020, the German Federal Supreme Court (BGH) rearmed this ruling

and ruled that non-essential cookies may not be activated on websites

without the user’s consent.

e major browser providers Apple with Safari and Mozilla with

Firefox have already rejected third-party cookies. Google wants to follow

suit with Chrome 2022. is poses dramatic challenges, especially for

performance-based business models: Agencies, data traders, media opera-

tors and marketers will lose an important basis for their business.

But also all companies that want to address their customers more

specically by means of data are aected by this. Following a user through

his specic customer journey is no longer possible. Digital campaigns can

then only be implemented on the remaining data-based models and in

It states in recital 30: “Natural persons may be associated with online identiers such as IP

addresses and cookie identiers provided by his or her device or software applications and tools or

logs, or other identiers such as radio frequency identiers. is may leave traces that can be used,

especially in combination with unique identiers and other information received by the server, to

prole and identify individuals.” (https://dsgvo-gesetz.de/erwaegungsgruende/nr-30/)

H.-W. Eckert

135

the environment planning with the classic advertising media. e companies

that will benet most from this are those that have already collected a

large amount of their own data, especially Amazon, Facebook, and Google.

is weakens the position of online retailers, e-commerce platforms

and all other companies that use digital channels to contact their target

groups. Personalized communication will rely more on predictions and

will have to develop appropriate models. e industry is in the process of

developing alternative personalization models that do not require user

identication. Semantic techniques are increasingly being used in the

contextual analysis of advertising elds. is used to be called contextual

targeting, but has become more in-depth and meaningful with the

semantic analysis techniques shown in Sect. 3.3.4. Whereas in the past it

was certain keywords that appeared in a text, today text content can also

be evaluated statistically and linguistically. With these methods, content

can be analysed, meaning correlations can be established and linked to

advertising messages. To do this, signicantly larger volumes of data must

be analysed in order to be able to form patterns for addressing target

groups. Machine learning plays an important role here.

e exclusion of third parties forces a change of strategy in communi-

cation: away from the performance-focused approach with an eye on the

last click towards a holistic view of the customer relationship. It breaks

the xation on the end of the sales funnel with one-to-one personalisa-

tion and creates space again for the development of customer groups and

a process of permanent optimisation of the approach. e focus is then

less on the channels and more on the interests and needs of the custom-

ers. All channels in the customer approach can thus be coordinated and

mapped in a dramaturgy. In the future, this will lead away from rule-

based marketing automation (with classic if-then rules) to an orchestra-

tion of all contact points and communication channels.

However, this is only possible if the company’s own data is consistently

collected and linked with one another. Communication managers are

only on the safe side if they build up data records from their own sources

(i.e. rst-party data). is requires a planned approach, as this data can-

not be bought, but has to be developed by the company itself. Knowledge

about customers is available in many places in the company (especially

sales, service) and stored in some systems (especially ERP and CRM). It

5 Fair Play: What Counts in Data Stories

136

can also be collected and further enriched in the many contact points of

the communication units (especially with data from the websites, social

media channels and email marketing). Above all, it is important to have

a common understanding of the objectives to be achieved by collecting

and using the data. Restricting the use of third-party data is a good

opportunity to take another general look at the reasons for building and

expanding one’s own data base and to start by looking for the questions

that are to be answered by using the data. To use the image used at the

beginning: ere are enough sources of one’s own that are worth develop-

ing and processing. For the statements of the oracle, however, it is not the

quantity of smoke that decides, but the quality of the question and the

abilities of the team of pythia and priests that delivers the answers. e

latter should be aware that not only their own thought patterns are sus-

ceptible to distortion, but also the tools they use to process the data.

References

Angwin J, Larson J, Mattu S, Kirchner L (2016) Machine bias. ProPublica (23

Mai) https://www.propublica.org/article/machine- bias- risk- assessments- in-

criminal- sentencing. Accessed: 23. Mai 2020

Balsillie J (2018) Data is not the new oil– it’s the new plutonium. Financial

Post, 18 Mai. https://business.nancialpost.com/technology/jim- balsillie-

data- is- not- the- new- oil- its- the- new- plutonium. Accessed: 22. Jan. 2020

Barocas S, Selbst A (2016) Big data’s disparate impact. 104 California Law

Review 671. https://ssrn.com/abstract=2477899. Accessed: 10. Aug. 2020

Bitkom (2018) Empfehlungen für den verantwortlichen Einsatz von KI und

automatisierten Entscheidungen, Berlin

Burack C (2017) Algorithms are ‘existential threat’ to shared reality, says Cathy

O’Neil, in: DW 21.08.2018 https://www.dw.com/en/algorithms- are-

existential- threat- to- shared- reality- says- cathy- oneil/a- 40167802. Accessed:

2. Mai 2020

Bynum T (2018) Computer and information ethics. e Stanford encyclopedia

of philosophy, HG. Edward N. Zalta. https://plato.stanford.edu/archives/

sum2018/entries/ethics- computer. Accessed: 20. Aug. 2020

Deloitte (2020) State of AI in the enterprise– 3rd Edition. Deloitte, Juni 2020

H.-W. Eckert

137

Economist (2019) Companies should take California’s new data-privacy law

seriously. Economist, 18. Dezember. https://www.economist.com/busi-

ness/2019/12/18/companies- should- take- californias- new- data- privacy- law-

seriously Accessed: 23. Dez. 2019

Economist (2020) Governments are erecting borders for data. Economist, 20.

Februar. https://www.economist.com/special- report/2020/02/20/governments-

are- erecting- borders- for- data. Accessed: 23. Febr. 2020

Eichsteller H, Seitz J (2019) Digital dialog insights 2019. Bundesanzeiger, Köln

Europäische Kommission (2020) Europäische Datenstrategie. Website der

Europäischen Kommission, 19. Februar. https://ec.europa.eu/info/strategy/

priorities- 2019- 2024/europe- fit- digital- age/european- data- strategy_de.

Accessed: 21. Aug. 2020

European Parliament (2019) A governance framework for algorithmic account-

ability and transparency, EPRS Brüssel. https://www.europarl.europa.eu/

RegData/etudes/STUD/2019/624262/EPRS_STU(2019)624262_EN.pdf

Greussing E (2019) Datenvisualisierung vom Publikum her denken. European

Journalism Observatory. https://de.ejo- online.eu/digitales/datenvisuali

sierung- vom- publikum- her- denken. Accessed: 21. Aug. 2020

Harari Y (2017) Homo deus: a brief history of tomorrow. Harper, NewYork

Jonauskaite D etal (2020) Universal patterns in color-emotion associations are

further shaped by linguistic and geographic proximity. Psychol Sci. https://

doi.org/10.1177/0956797620948810. Accessed: 29. Sept. 2020

Kahneman D (2011) inking, fast and slow. Penguin, London

Kaltheuner F, Obermüller N (2018) Diskriminierende Gesichtserkennung: Ich

sehe was, was du nicht bist. Netzpolitik.org. https://netzpolitik.org/2018/

diskriminierende- gesichtserkennung- ich- sehe- was- was- du- nicht- bist/.

Accessed: 9. Aug. 2020

Kutsche K (2019) Wenn Daten in die Irre führen. Süddeutsche Zeitung, 17.

Dezember. https://www.sueddeutsche.de/wirtschaft/kuenstliche- intelligenz-

wenn- daten- in- die- irre- fuehren- 1.4726877. Accessed: Sept. 2020

Lobe A (2019) Speichern und Strafen. Die Gesellschaft im Datengefängnis,

Beck, München

Matzat L (2017) Algorithmic accountability. Der nächste Schritt für den

Datenjournalismus Datenjournalist, 18. Dezember. https://www.datenjour-

nalist.de/algorithmic- accountability- der- naechste- schritt- fuer- den-

datenjournalismus/. Accessed: 23. Mai 2020

O’Neil C (2017) Weapons of math destruction. Penguin Books, London

Rosling H (2018) Factfulness. Sceptre, London

5 Fair Play: What Counts in Data Stories

138

Schneider B (2020) „Wir können nicht die ganze Zeit hinschauen“, Süddeutsche

Zeitung Magazin. https://zeitung.sueddeutsche.de/webapp/issue/SZM/

2020- 10/8/index.html. Accessed: 10. März 2020

Schüller K, Busch P, Hindinger C (2019) Future Skills: Ein Framework für Data

Literacy, Hochschulforum Digitalisierung Arbeitspapier Nr. 47. Berlin

Sedikides C, Skowronski J (2020) In human memory. Good can be stronger

than bad. In: current directions in psychological science 29/1, S 86–91

Weizenbaum J (1976) Computer power and human reason: from judgment to

calculation. Freeman, San Francisco

Wiener N (1948) Cybernetics. MIT Press, Cambridge

Wilkinson MD et al (2016) e FAIR guiding principles for scientic data

management and stewardship. Scientic Data 3:160018. https://www.ncbi.

nlm.nih.gov/pmc/articles/PMC4792175/. Accessed: 21. Aug. 2020

H.-W. Eckert